Audio decoder and audio decoding method

Abstract

First determiner 121 provisionally determines whether a current processing unit is a stationary noise region based on a determination result on stationary characteristics of a decoded signal. Based on the provisional determination result and a determination result on periodicity of the decoded signal, second determiner 124 further determines whether the current processing unit is a stationary noise region, whereby a decoded signal including a stationary speech signal such as a stationary vowel is distinguished from a stationary noise, and thus the stationary noise region is detected accurately.

Description

TECHNICAL FIELD

[0001] The present invention relates to a speech decoding apparatus that decodes speech signals encoded at a low bit rate in a mobile communication system and packet communication system including internet communications where the speech signals are encoded and transmitted, and more particularly, to a CELP (Code Excited Linear Prediction) speech decoding apparatus that divides the speech signals to spectral envelope components and residual components to represent.

BACKGROUND ART

[0002] In fields of digital mobile communications, packet communications as typified by internet communications and speech storage, speech coding apparatuses are used which compress speech information to effectively use the capacity of transmission path of radio signals and storage media to encode with high efficiency. Among those, systems based on CELP (Code Excited Linear Prediction) system are carried into practice widely at medium and low bit rates. Techniques of CELP are described in M. R. Schroeder and B. S. Atal: “Code-Excited Linear Prediction (CELP): High-quality Speech at Very Low Bit Rates”, Proc. ICASSP-85,25.1.1, pages 937-940, 1985.

[0003] In the CELP speech coding system, a speech is divided into frames each with a constant length (about 5 ms to 50 ms), linear prediction analysis is performed for each frame, a prediction residual (excitation signal) by linear prediction for each frame is encoded using an adaptive code vector and fixed code vector each composed of a known waveform. The adaptive code vector is selected from an adaptive codebook that stores excitation vectors previously generated, and the fixed code vector is selected from a fixed codebook that stores a predetermined number of beforehand prepared vectors with predetermined shapes. As fixed code vectors stored in the fixed codebook are used random vectors and vectors generated by arranging a number of pulses at different positions.

[0004] A conventional CELP coding apparatus performs analysis and quantization of LPC (Liner Predictive Coefficient), pitch search, fixed codebook search and gain codebook search using input digital signals, and transmits LPC code (L), pitch period (A), fixed codebook index (F) and gain codebook index (G) to a decoding apparatus.

[0005] The decoding apparatus decodes LPC code (L), pitch period (A), fixed codebook index (F) and gain codebook index (G), and based on the decoding results, drives a synthesis filter with the excitation signal to obtain a decoded speech.

[0006] However, in the conventional speech decoding apparatus, it is difficult to detect a stationary noise region by distinguishing signals such as stationary vowels that are stationary but are not noises from stationary noises.

DISCLOSURE OF INVENTION

[0007] It is an object of the present invention to provide a speech decoding apparatus that detects stationary noise signal regions accurately to decode speech signals, specifically, a speech decoding apparatus and speech decoding method which enable determination of speech region or non-speech region, distinguish a periodical stationary signal from a stationary noise signal like a white noise using a pitch period and adaptive code gain, and detect a stationary noise signal region accurately.

[0008] The object is achieved by provisionally determining stationary noise characteristics of a decoded signal, further determining whether a current processing unit is a stationary noise region based on the provisional determination result and a determination result on the periodicity of the decoded signal, distinguishing the decoded signal containing a stationary speech signal such as a stationary vowel from a stationary noise, and detecting the stationary noise region properly.

BRIEF DESCRIPTION OF DRAWINGS

[0009] FIG. 1 is a diagram illustrating a configuration of a stationary noise region determining apparatus according to a first embodiment of the present invention;

[0010] FIG. 2 is a flow diagram illustrating procedures of grouping of pitch history;

[0011] FIG. 3 is a diagram illustrating part of the flow of mode selection:

[0012] FIG. 4 is another diagram illustrating part of the flow of mode selection:

[0013] FIG. 5 is a diagram illustrating a configuration of a stationary noise post-processing apparatus according to a second embodiment of the present invention;

[0014] FIG. 6 is a diagram illustrating a configuration of a stationary noise post-processing apparatus according to a third embodiment of the present invention;

[0015] FIG. 7 is a diagram illustrating a speech decoding processing system according to a fourth embodiment of the present invention;

[0016] FIG. 8 is a flow diagram illustrating the flow of the speech decoding system;

[0017] FIG. 9 is a diagram illustrating examples of memories provided in the speech decoding system and of initial values of the memories;

[0018] FIG. 10 is a diagram illustrating the flow of mode determination processing;

[0019] FIG. 11 is a diagram illustrating the flow of stationary noise addition processing; and

[0020] FIG. 12 is a diagram illustrating the flow of scaling.

BEST MODE FOR CARRYING OUT THE INVENTION

[0021] Embodiments of the present invention will be described below with reference to accompanying drawings.

[0022] (First Embodiment)

[0023] FIG. 1 illustrates a configuration of a stationary noise region determining apparatus according to the first embodiment of the present invention.

[0024] A coder (not shown) first performs analysis and quantization of LPC (Liner Prediction Coefficients), pitch search, fixed codebook search and gain codebook search using input digital signals, and transmits LPC code (L), pitch period (A), fixed codebook index (F) and gain codebook index (G).

[0025] Code receiving apparatus 100 receives a coded signal transmitted from the coder, and divides code L representing LPC, code A representing an adaptive code vector, code G representing gain information and code F representing a fixed code vector from the received signal. The divided code L, code A, code G and code F are output to speech decoding apparatus 101. Specifically, code L is output to LPC decoder 110, code A is output to adaptive codebook 111, code G is output to gain codebook 112, and code F is output to fixed codebook 113.

[0026] Speech decoding apparatus 101 will be described first.

[0027] LPC decoder 110 decodes LPC from code L to output to synthesis filter 117. LPC decoder 110 converts the decoded LPC into LSP (Line Spectrum Pairs) parameter to exploit their better interpolation property, and outputs LSP to inter-subframe variation calculator 119, distance calculator 120 and average LSP calculator 125 provided in stationary noise region detecting apparatus 102.

[0028] In general, LPC are coded in LSP domain, i.e. code L is coded LSP, and in the cases, the LPC decoder decodes LSP and then converts the decoded LSP to LPC. LSP parameter is one of examples of spectral envelope parameters representing a spectral envelope component of a speech signal. The spectral envelope parameters include PARCOR coefficient or LPC.

[0029] Adaptive codebook 111 provided in speech decoding apparatus 101 updates previously generated excitation signals to temporarily store as a buffer, and generates an adaptive code vector using an adaptive codebook index (pitch period (pitch lag)) obtained by decoding input code A. The adaptive code vector generated in adaptive codebook 111 is multiplied by an adaptive code gain in adaptive code gain multiplier 114 and then output to adder 116. The pitch period obtained in adaptive codebook 111 is output to pitch history analyzer 122 provided in stationary noise region detecting section 102.

[0030] Gain codebook 112 stores a predetermined number of sets (gain vectors) of adaptive codebook gain and fixed codebook gain, and outputs an adaptive codebook gain component (adaptive code gain) to adaptive code gain multiplier 114 and second determiner 124, and further outputs a fixed codebook gain component (fixed code gain) to fixed code gain multiplier 115, where the components are of a gain vector designated by a gain codebook index obtained by decoding input code G.

[0031] Fixed codebook 113 stores a predetermined number of fixed code vectors with different shapes, and outputs a fixed code vector designated by a fixed codebook index obtained by decoding input code F to fixed code gain multiplier 115. Fixed code gain multiplier 115 multiplies the fixed code vector by the fixed code gain to output to adder 116.

[0032] Adder 116 adds the adaptive code vector input from adaptive code gain multiplier 114 and the fixed code vector input from fixed code gain multiplier 115 to generate an excitation signal for synthesis filter 117, and outputs the signal to synthesis filter 117 and adaptive codebook 111.

[0033] Synthesis filter 117 constructs an LPC synthesis filter using LPC input from LPC decoder 110. Synthesis filter 117 performs filtering processing using the excitation signal input from adder 116 as an input to synthesize a decoded speech signal, and outputs the synthesized decoded speech signal to post filter 118.

[0034] Post filter 118 performs processing such as formant enhancement and pitch enhancement to improve the subjective quality on the synthesized signal output from synthesis filter 117. The speech signal subjected to the processing is output to as a final post-filter output signal of speech decoding apparatus 101 to power variation calculator 123 provided in stationary noise region detecting apparatus 102.

[0035] The decoding processing in speech decoding apparatus 101 as described above is executed on a processing unit with a predetermined time (frame of a few tens of milliseconds) basis or on a processing unit (subframe) divided from a frame basis. A case will be described below where processing is executed on a subframe basis.

[0036] Stationary noise region detecting apparatus 102 will be described below. First stationary noise region detecting section 103 provided in stationary noise region detecting apparatus 102 is first explained. First stationary noise region detecting section 103 and second stationary noise region detecting section 104 perform mode selection and determines whether a subframe is a stationary noise region or speech signal region.

[0037] LSP output from LPC decoder 110 is output to first stationary noise region detecting section 103 and stationary noise characteristic extracting section 105 provided in stationary noise region detecting apparatus 102. LSP input to first stationary noise region detecting section 103 is input to inter-subframe variation calculator 119 and distance calculator 120.

[0038] Inter-subframe variation calculator 119 calculates a variation in LSP from an immediately preceding (last) subframe. Specifically, based on LSP input from LPC decoder 110, the calculator 119 calculates a difference in LSP between a current subframe and last subframe for each order, and outputs the square sum of the differences as an inter-subframe variation amount to first determiner 121 and second determiner 124.

[0039] In addition, it is preferable to use smoothed version of LSP in calculating the variation amount, for reducing effects of the fluctuations of quantization error and so on. Strong smoothing causes too slow variations between subframes, and therefore, the smoothing is set to be weak. For example, when smoothing LSP is defined as expressed in (Eq.1), it is preferable to set k at about 0.7.

Smoothing LSP [current subframe]=k×LSP+(1−k)×smoothing LSP [last subframe]  (Eq.1)

[0040] Distance calculator 120 calculates a distance between average LSP in a previous stationary noise region input from average LSP calculator 125 and LSP of the current subframe input from LPC decoder 110, and outputs the calculation result to first determiner 121. As the distance between average LSP and LSP of the current subframe, for example, distance calculator 120 calculates for each order a difference between average LSP input from average LSP calculator 125 and LSP of the current subframe input from LPC decoder 110, and outputs the square sum of the differences. Distance calculator 120 may output the differences in LSP calculated for each order without square summing. Further, in addition to these values, the calculator 120 may outputs a maximum value of the differences in LSP calculated for each order. Thus, by outputting various measures of distance to first determiner 121, it is possible to improve determination accuracy in first determiner 121.

[0041] Based on the information input from inter-subframe variation calculator 119 and distance calculator 120, first determiner 121 determines a degree of the variation in LSP between subframes, and a similarity (distance) between LSP of the current subframe and average LSP of the stationary noise region. Specifically, these determinations are made using threshold processing. When it is determined that the variation in LSP between subframes is small and LSP of the current subframe is similar to average LSP of the stationary noise region (i.e. the distance is small), the current subframe is determined as a stationary noise region. The determination result (first determination result) is output to second determiner 124.

[0042] In this way, first determiner 121 provisionally determines whether a current subframe is a stationary noise region. This determination is made by determining stationary characteristics of a current subframe based on a variation amount in LSP between the last subframe and current subframe, and further determining noise characteristics of the current subframe based on the distance between average LSP and LSP of the current subframe.

[0043] However, the determination based on only LSP sometimes erroneously determines that a periodical stationary signal such as a stationary vowel or sine wave is a noise signal. Therefore, second determiner 124 provided in second stationary noise region detecting section 104 as described below analyzes the periodicity of the current subframe, and based on the analysis result, determines whether the current subframe is a stationary noise region. In other words, since a signal with high periodicity has a high possibility of being a stationary vowel or the like (i.e. not noise), second determiner 124 determines such a signal is not a stationary noise region.

[0044] Second stationary noise region detecting section 104 will be described below.

[0045] Pitch history analyzer 122 analyzes fluctuations between subframes in pitch period input from the adaptive codebook. Specifically, pitch history analyzer 122 temporarily stores pitch periods input from adaptive codebook 111 corresponding to a predetermined number of subframes (for example, ten subframes), and performs grouping on the temporarily stored pitch periods (pitch periods of last ten subframes including the current subframe) by the method as illustrated in FIG. 2.

[0046] The grouping will be described using as an example a case of performing grouping on pitch periods of last ten subframes including a current subframe. FIG. 2 is a flow diagram illustrating procedures of performing the grouping. First, in ST1001, pitch periods are classified. Specifically, pitch periods with the same value are sorted into a same class. In other words, pitch periods with the exactly same value are sorted into a same class, while a pitch period with even a little different value is sorted into a different class.

[0047] Next, in ST1002, among classified classes, grouping is performed that classes having close pitch period values are grouped into a single group. For example, classes with pitch periods between which differences are within 1 are sorted into a single group. In performing the grouping, when there are five classes where mutual differences in pitch period are within 1 (for example, classes with pitch periods respectively of 30, 31, 32, 33 and 34), the five classes may be sorted into a single group.

[0048] In ST1003, as a result of the grouping, a result of the analysis is output that indicates the number of groups to which pitch periods in last ten subframes including the current subframe belong. As the number of groups indicated by the result of the analysis is decreased, the possibility is increased that the decoded speech signal is periodical, while as the number of groups is increased, the possibility is increased that the decoded speech signal is not periodical. Accordingly, when the decoded speech signal is stationary, it is possible to use the result of the analysis as a parameter indicative of periodical stationary signal characteristics (periodicity of a stationary noise).

[0049] Power variation calculator 123 receives as its inputs the post-filter output signal input from post filter 118 and average power information of the stationary noise region input from average noise power calculator 126. Power variation calculator 123 obtains the power of the post-filter output signal input from post filter 118, and calculates the ratio (power ratio) of the obtained power of the post-filter output signal to the average power of the stationary noise region. The power ratio is output to second determiner 124 and average noise power calculator 126. The power information of the post-filter output signal is also output to average noise power calculator 126. When the power (current signal power) of the post-filter output signal output from post filter 118 is larger than the average power of the stationary noise region, there is a possibility that the current subframe is a speech region. The average power of the stationary noise region and the power of the post-filter output signal output from post filter 118 are used as parameters to detect, for example, onset regions of a speech that is not detected using other parameters. In addition, power variation calculator 123 may calculate a difference in the power to use as a parameter, instead of the ratio of the power of the post-filter output signal to the average power of the stationary noise region.

[0050] As described above, to second determiner 124 are input pitch history analysis result (the number of groups) in pitch history analyzer 122 and the adaptive code gain obtained in gain codebook 112. Using the input information, second determiner 124 determines the periodicity of the post-filter output signal. To second determiner 124 are further input the first determination result in first determiner 121, the ratio of the power of the current subframe to the average power of the stationary noise region calculated in power variation calculator 123, and the inter-subframe variation amount in LSP calculated in inter-subframe variation calculator 119. Based on the input information, the first determination result, and the determination result on the above-mentioned periodicity, second determiner 124 determines whether the current subframe is a stationary noise region, and outputs the determination result to a processing apparatus provided downstream. The determination result is also output to average LSP calculator 125 and average noise power calculator 126. In addition, it may be possible to provide either code receiving apparatus 100, speech decoding apparatus 101 or stationary noise region detecting apparatus 102 with a decoding section that decodes information indicative of whether a state is a speech stationary state contained in the received coded, and outputs the information indicative of whether a state is a speech stationary state to second determiner 124.

[0051] Stationary noise characteristic extracting section 105 will be described below.

[0052] Average LSP calculator 125 receives as its inputs the determination result from second determiner 124, and LSP of the current subframe from speech decoding apparatus 101 (more specifically, LPC decoder 110). Only when the determination result indicates a stationary noise region, average LSP calculator 125 updates the average LSP in the stationary noise region using the input LSP of the current subframe. The average LSP is updated, for example, using the AR smoothing equation. The updated average LSP is output to distance calculator 120.

[0053] Average noise power calculator 126 receives as its inputs the determination result from second determiner 124, and the power of the post-filter output signal and the power ratio (the power of the post-filter output signal the average power of the stationary noise region) from power variation calculator 123. In the case where the determination result from second determiner 124 indicates a stationary noise region, and in the case where (the determination result does not indicate a stationary noise region, but) the power ratio is smaller than a predetermined threshold (the power of the post-filter output signal of the current subframe is smaller than the average power of the stationary noise region), average noise power calculator 126 updates the average power (average noise power) of the stationary noise region using the input post-filter output signal power. The average noise power is updated, for example, using the AR smoothing equation. In this case, by adding control of decreasing the smoothing as the power ratio is decreased (so that the post-filter output signal power of the current subframe tends to be reflected), it is possible to decrease a level of the average noise power promptly even when the background noise level decreases rapidly in a speech region. The updated average noise power is output to power variation calculator 123.

[0054] In the above-mentioned configuration, LPC, LSP and average LSP are parameters indicative of a spectral envelope component of a speech signal, while the adaptive code vector, noise code vector, adaptive code gain and noise code gain are parameters indicative of a residual component of the speech signal. Parameters indicative of a spectral envelope component and parameters indicative of a residual component are not limited to the above-mentioned information.

[0055] Procedures of the processing will be described below in first determiner 121, second determiner 124, and stationary noise characteristic extracting section 105 with reference to FIGS. 3 and 4. In FIGS. 3 and 4, processing of ST1101 to ST1107 is principally performed in first stationary noise region detecting section 103, processing of ST1108 to ST1117 is principally performed in second stationary noise region detecting section 104, and processing of ST1118 to ST1120 is principally performed in stationary noise characteristic extracting section 105.

[0056] In ST1101, LSP of a current subframe is calculated, and the calculated LSP undergoes the smoothing as expressed by (Eq.1) as described previously. In ST1102, a difference (variation amount) in LSP between the current subframe and the last (immediately preceding) subframe is calculated. The processing of ST1101 and ST1102 is performed in inter-subframe variation calculator 119 as described previously.

[0057] An example of the method of calculating the variation amount in LSP in inter-subframe variation calculator 119 is indicated in (Eq.1′), (Eq.2) and (Eq.3). (Eq.1′) is an equation to perform smoothing on LSP of the current subframe, (Eq.2) is an equation to calculate the square sum of differences in LSP subjected to the smoothing between subframes, and (Eq.3) is an equation to further perform smoothing on the square sum of differences in LSP between subframes. L′i (t) represents an ith-order smoothed LSP parameter in a tth subframe, Li (t) represents an ith-order LSP parameter in the tth subframe, DL(t) represents an LSP variation amount (the square sum of differences between subframes) in the tth subframe, DL′(t) represents a smoothed version of LSP variation amount in the tth subframe, and p represents a LSP (LPC) analysis order. In this example, inter-subframe variation calculator 119 obtains DL′(t) using (Eq.1′), (Eq.2) and (Eq.3), and the obtained DL′(t) is used as the inter-subframe variation amount in LSP in mode determination.

L′i (t)=0.7×Li(t)+0.3×L′i(t−1)  (Eq.1′) 1 $$ DL ⁡ ( t ) = ∑ i = 1 p ⁢   ⁢ { [ L ′ ⁢ i ⁡ ( t ) - L ′ ⁢ i ⁡ ( t - 1 ) ] 2 } ( Eq .   ⁢ 2 )

DL ′(t)=0.1×DL(t)+0.9×DL′(t−1)  (Eq.3)

[0058] In ST1103, distance calculator 120 calculates a distance between LSP of the current subframe and average LSP in the previous noise region. (Eq.4) and (Eq.5) indicate a specific example of distance calculation in distance calculator 120. (Eq.4) defines the distance between the average LSP in the previous noise region and LSP of the current subframe as the square sum of differences of all the orders, and (Eq.5) defines the distance as the square of only a difference of the order where the difference is the largest. LNi is the average LSP in the previous noise region, and is updated in a noise region, for example, using (Eq.6) on a subframe basis. In this example, distance calculator 120 obtains D(t) and DX(t) using (Eq.4), (Eq.5) and (Eq.6), and obtained D(t) and DX(t) are used as information of the distance from LSP of the stationary noise region in mode determination. 2$\begin{array}{cc}D\left(t\right)=\sum _{i=1}^p\left\{{\left[\mathrm{Li}\left(t\right)-\mathrm{LNi}\right]}^2\right\}& \left(\mathrm{Eq}.4\right)\end{array}$

DX (t)=Max{[Li(t)−LNi]2}i=1,,,p  (Eq.5)

LNi= 0.95×LNi+0.05×Li(t)  (Eq.6)

[0059] In ST1104, power variation calculator 123 calculates the power of the post-filter output signal (output signal from post filter 118). The calculation of the power is performed in power variation calculator 123 as described previously, and more specifically, the power is obtained using (Eq.7), for example. In (Eq.7), S(i) is the post-filter output signal, and N is the length of a subframe. Since the power calculation in ST1104 is performed in power variation calculator 123 provided in second stationary noise region detecting section 104 as illustrated in FIG. 1, it is only required to perform the power calculation prior to ST1108, and the timing of power calculation is not limited to a position of ST1104. 3$\begin{array}{cc}P=\sqrt{\left\{\sum _{i=0}^N\left[S\left(i\right)\times S\left(i\right)\right]\right\}}& \left(\mathrm{Eq}.7\right)\end{array}$

[0060] In ST 1105, determination is made on stationary noise characteristics of a decoded signal. Specifically, it is determined whether the variation amount calculated in ST 1102 is small in value and the distance calculated in ST 1103 is small in value. In other words, a threshold is set with respect to each of the variation amount calculated in ST1102 and distance calculated in ST1103, and when the variation amount calculated in ST1102 is smaller than the set threshold and the distance calculated in ST1103 is also smaller than the set threshold, the stationary noise characteristics are high and the processing flow shifts to ST1107. For example, with respect to DL′D and DX as described previously, when LSP is normalized in a range of 0.0 to 1.0, using thresholds as described below enables the determination with high accuracy.

[0061] Threshold for DL: 0.0004

[0062] Threshold for D: 0.003+D′

[0063] Threshold for DX: 0.0015

[0064] D′ is an average value of D in a noise region, and for example, is calculated using (Eq.8) in a noise region.

D′= 0.05×D(t)+0.95×D′  (Eq.8)

[0065] Since LNi that is the average LSP in the previous noise region has an adequately reliable value only when the noise region with a sufficient time somewhat (for example, corresponding to about 20 subframes) is available, D and DX are not used in the determination on stationary noise characteristics in ST1005 when the previous noise region is smaller than a predetermined time length (for example, 20 subframes).

[0066] In ST1107, the current subframe is determined as a stationary noise region, and the processing flow shifts to ST1108. Meanwhile, when either the variation calculated in ST1102 or the distance calculated in ST1103 is larger than the threshold, the current subframe is determined to have low stationary characteristics and the processing flow shifts to ST1106. In ST1106, it is determined that the subframe is not a stationary noise region (in other words, speech region), and the processing flow shifts to ST1110.

[0067] In ST1108, it is determined whether the power of the current subframe is larger than the average power of the pervious stationary noise region. Specifically, a threshold is set with respect to an output result of power variation calculator 123 (the ratio of the power of the post-filter output signal to the average power of the stationary noise region), and when the ratio of the power of the post-filter output signal to the average power of the stationary noise region is larger than the set threshold, the processing flow shifts to ST1109, and in ST1109 the current subframe is corrected in determination to be a speech region.

[0068] As a specific value of the threshold using 2.0 (i.e. the processing flow shifts to ST1109 when the power P of the post-filter output signal obtained using (Eq.7) exceeds twice the average power PN′ of the stationary noise region obtained in the noise region, average power PN′ is updated for each subframe during the stationary noise region, for example, using (Eq.9)) enables the determination with high accuracy.

PN′= 0.9×PN′+0.1×P  (Eq.9)

[0069] Meanwhile, in the case where the power variation is smaller than the set threshold, the processing flow shifts to ST1112. In this case, the determination result in ST1107 is kept, and the current subframe is still determined as a stationary noise region.

[0070] Next, in ST1110, it is checked how long the stationary state lasts and whether the stationary state is a stationary voiced speech. Then, when the current subframe is not a stationary voiced speech and the stationary state has lasted for a predetermined time duration, the processing flow proceeds to ST1111, and in ST1111 the current subframe is re-determined as a stationary noise region.

[0071] Specifically, whether the current subframe is in a stationary state is determined using the output (inter-subframe variation amount) of inter-subframe variation calculator 119. In other words, when the inter-subframe variation amount obtained in ST1102 is small (smaller than the predetermined threshold (for example, the same value as the threshold used in ST1105)), the current subframe is determined as the stationary state. Thus, when the stationary noise state is determined, it is checked how long the state has lasted.

[0072] The check on whether the current subframe is a stationary voiced speech is performed based on information indicative of whether the current subframe is the stationary voiced speech provided from stationary noise region detecting apparatus 102. For example, when the transmitted code information includes such information as the mode information, it is check whether the current subframe is a stationary voiced speech, using the decoded mode information. Otherwise, a section that determines speech stationary characteristics provided in stationary noise region detecting apparatus 102 outputs such information, and using the information, the stationary voiced speech is checked.

[0073] As a result of the check, in the case where the stationary state has lasted for a predetermined time duration (for example, 20 subframes or more) and is not the stationary voiced speech, the current subframe is re-determined as a stationary noise region in ST1111 and the processing flow shifts to ST1112 even when it is determined that the power variation is large in ST1108. On the other hand, when the determination result in ST1110 is “No” (a case of speech stationary region or a case where a stationary state has not lasted for a predetermined time duration), the determination result that the current subframe is a speech region is kept and the processing flow shifts to ST1114.

[0074] Next, when it is determined that the current subframe is a stationary noise region in processes up to this point, whether the periodicity of the decoded signal is high is determined in ST1112. Specifically, based on the adaptive code gain input from speech decoding apparatus 101 (more specifically, gain codebook 112) and pitch history analysis result input from pitch history analyzer 122, second determiner 124 determines the periodicity of the decoded signal in the current subframe. In this case, as an adaptive code gain, it is preferable to use a smoothed version in order for the variation between subframes to be smoothed.

[0075] The determination on the periodicity is made, for example, by setting a threshold with respect to the smoothed adaptive code gain, and when the smoothed adaptive code gain exceeds the predetermined threshold, it is determined that the periodicity is high and the processing flow shifts to ST1113. In ST1113, the current subframe is re-determined as a speech region.

[0076] Further, since the possibility is higher that periodical signals are continued as the number of groups is smaller to which pitch periods in previous subframes belong in the pitch history analysis result, the periodicity is determined based on the number of groups. For example, when pitch periods of previous ten subframes are sorted into groups of three or less, since the possibility is high of a region where the periodical signal lasts, the processing flow shifts to ST1113, and the current subframe is re-determined to be a speech region (not a stationary noise region).

[0077] When the determination result in ST1112 indicates “No” (the smoothed adaptive code gain is smaller than the predetermined threshold and previous pitch periods are sorted into a large number of groups in the pitch history analysis result), the determination result indicative of the stationary noise region is maintained and the processing flow shifts to ST1115.

[0078] When the determination result indicates a speech region in processes up to this point, the processing flow shifts to ST1114 and a hangover counter is set for the predetermined number of hangover subframes (for example, 10). The hangover counter is set for the number of hangover frames as an initial value, and is decremented by 1 whenever a stationary noise region is determined according to the processing of ST1101 to ST1113. Then, when the hangover counter is “0”, the current subframe is finally determined as a stationary noise region in the method of determining a stationary noise region.

[0079] When the determination result indicates a noise stationary region in processes up to this point, the processing flow shifts to ST1115 and it is checked whether the hangover counter is within a hangover range (“1” to “the number of hangover frames”). In other words, it is checked whether the hangover counter is “0”. When the hangover counter is within the hangover range, (in a range from “1” to “the number of hangover frames”), the processing flow shifts to ST1116 where the determination result is corrected to be a speech region and the processing flow shifts to ST1117. In ST1117, the hangover counter is decremented by 1. When the counter is not in the hangover range (is “0”), the determination result indicative of a stationary noise region is maintained and the processing flow shifts to ST1118.

[0080] When the determination result indicates the stationary noise region, average LSP calculator 125 updates the average LSP in the stationary noise region in ST1118. The update is performed, for example, using (Eq.6) when the determination result indicates the stationary noise region, while the previous value is maintained without being updated when the determination result does not indicate the stationary noise region. In addition, when the time duration previously determined as a stationary noise region is short, the smoothing coefficient, 0.95, in (Eq.6) may be decreased.

[0081] In ST1119, average noise power calculator 126 updates the average noise power. The update is performed, for example, using (Eq.9) when the determination result indicates the stationary noise region, while the previous value is maintained without being updated when the determination result does not indicate the stationary noise region. However, when the determination result does not indicate the stationary noise region, but the power of the current post-filter output power is smaller than the average noise power, the average noise power is updated using the same equation as (Eq.9) except the smoothing coefficient that is smaller than 0.9 to decrease the average noise power. By performing such update, it is possible to handle the cases where the background noise level suddenly decreases during a speech region.

[0082] Finally, in ST1120, second determiner 124 outputs the determination result, average LSP calculator 125 outputs the updated average LSP, and average noise power calculator 126 outputs the updated average noise power.

[0083] As described above, according to this embodiment, even when it is determined that a current subframe is a stationary noise region by judging stationary characteristics using LSP, a degree of periodicity of the current subframe is examined (determined) using the adaptive code gain and pitch period, and based on the degree of periodicity, it is checked again whether the current subframe is a stationary noise region. Accordingly, it is possible to make an accurate determination on signals such as sine waves and stationary vowels that are stationary but not noises.

[0084] (Second Embodiment)

[0085] FIG. 5 illustrates a configuration of a stationary noise post-processing apparatus according to the second embodiment of the present invention. In FIG. 5, the same sections as in FIG. 1 are assigned the same reference numerals as in FIG. 1, and specific descriptions thereof are omitted.

[0086] Stationary noise post-processing apparatus 200 is comprised of noise generating section 201, adder 202 and scaling section 203. Stationary noise post-processing apparatus 200 adds in adder 202 a pseudo stationary noise signal generated in noise generating section 201 and a post-filter output signal from speech decoding apparatus 101, performs in scaling section 203 scaling on the post-filter output signal subjected to the addition to adjust the power, and outputs the post-processing-processed post-filter output signal.

[0087] Noise generating section 201 is comprised of excitation generator 210, synthesis filter 211, LSP/LPC converter 212, multiplier 213, multiplier 214 and gain adjuster 215. Scaling section 203 is comprised of scaling coefficient calculator 216, inter-subframe smoother 217, inter-sample smoother 218 and multiplier 219.

[0088] The operation of stationary noise post-processing apparatus 200 with the above-mentioned configuration will be described below.

[0089] Excitation generator 210 selects a fixed code vector at random from fixed codebook 113 provided in speech decoding apparatus 101, and based on the selected fixed code vector, generates a noise excitation signal to output to synthesis filter 211. A method of generating a noise excitation signal is not limited to a method of generating the signal based a fixed code vector selected from fixed codebook 113 provided in speech decoding apparatus 101, and it may be possible to determine a method judged as the most effective for each system in terms of computation amount, memory capacity and also characteristics of generated noise signals. Generally it is the most effective selecting fixed code vectors from fixed codebook 113 provided in speech decoding apparatus 101. LSP/LPC converter 212 converts the average LSP from average LSP calculator 125 into LPC to output to synthesis filter 211.

[0090] Synthesis filter 211 constructs an LPC synthesis filter using LPC input from LSP/LPC converter 212. Synthesis filter 211 performs filtering processing using the noise excitation signal input from excitation generator 210 as its input to synthesize a noise signal, and outputs the synthesized noise signal to multiplier 213 and gain adjuster 215.

[0091] Gain adjuster 215 calculates a gain adjustment coefficient to scale up the power of the output signal of synthesis filter 211 to the average noise power from average noise power calculator 126. The gain adjustment coefficient undergoes the smoothing processing so that the smoothed continuity is maintained between subframes, and further undergoes the smoothing processing for each sample so that the smoothed continuity is maintained also in a subframe. Finally, a gain adjustment coefficient for each sample is output to multiplier 213. Specifically, the gain adjustment coefficient is obtained according to (Eq.10) to (Eq.12). Psn is the power of a noise signal synthesized in synthesis filter 211 (obtained in the same way as in (Eq.7)), and Psn′ is obtained by performing smoothing on Psn between subframes and is updated using (Eq.10). PN′ is the power of the stationary noise signal obtained in (Eq.9), and Scl is a scaling coefficient in a processing frame. Scl′ is a gain adjustment coefficient adopted for each sample, and is updated for each sample using (Eq.12).

Psn′= 0.9×Psn′+0.1×Psn  (Eq.10)

Scl=PN′/Psn′   (Eq.11)

Scl′= 0.85×Scl′+0.15×Scl  (Eq.12)

[0092] Multiplier 213 multiplies the gain adjustment coefficient input from gain adjuster 215 by the noise signal output from synthesis filter 211. The gain adjustment coefficient is variable for each sample. The multiplication result is output to multiplier 214.

[0093] In order to adjust an absolute level of a noise signal to generate, multiplier 214 multiplies a predetermined constant (for example, about 0.5) by the output signal from multiplier 213. Multiplier 214 maybe incorporated into multiplier 213. The level-adjusted signal (stationary noise signal) is output to adder 202. As described above, the stationary noise signal where the smoothed continuity is maintained is generated.

[0094] Adder 202 adds the stationary noise signal generated in noise generating section 201 to the post-filter output signal output from speech decoding apparatus 101 (more specifically, post filter 118) to output to scaling section 203 (more specifically, scaling coefficient calculator 216 and multiplier 219).

[0095] Scaling coefficient calculator 216 calculates both the power of the post-filter output signal output from speech decoding apparatus 101 (more specifically, post filter 118) and the power of the post-filter output signal to which the stationary noise signal added output from adder 202, calculates a ratio between both the power, and thus calculates a scaling coefficient for decreasing a variation in power between the scaled signal and decoded signal (to which the stationary noise is not added yet) to output to inter-subframe smoother 217. Specifically, the scaling coefficient SCALE is obtained as expressed by (Eq.13). P is the power of the post-filter output signal and is obtained in (Eq.7), and P′ is the power of the post-filter output signal to which the stationary noise signal is added and is obtained in the same equation as in P.

SCALE=P/P′  (Eq.13)

[0096] Inter-subframe smoother 217 performs the inter-subframe smoothing processing on the scaling coefficient so that the scaling coefficient varies gently between subframes. Such smoothing is not executed in a speech region (or extremely weak smoothing is executed). Whether a current subframe is a speech region is determined based on the determination result output from second determiner 124 as shown in FIG. 1. The smoothed scaling coefficient is output to inter-sample smoother 218. The smoothed scaling coefficient SCALE′ is updated by (Eq.14).

SCALE′=0.9×SCALE′+0.1×SCALE  (Eq.14)

[0097] Inter-sample smoother 218 performs the inter-sample smoothing processing on the scaling coefficient so that the scaling coefficient smoothed between subframes varies gently between samples. The smoothing processing can be performed by AR smoothing processing. Specifically, smoothed scaling coefficient SCALE for each sample is updated by (Eq.15).

SCALE″=0.85×SCALE″+0.15×SCALE′  (Eq.15)

[0098] In this way, the scaling coefficient is subjected to the smoothing processing between samples, and thus is varied gently for each sample, and it is thereby possible to prevent the scaling coefficient from being discontinuous near a boundary between subframes. The scaling coefficient calculated for each sample is output to multiplier 219.

[0099] Multiplier 219 multiplies the scaling coefficient output from inter-sample smoother 218 by the post-filter output signal to which the stationary noise signal is added input from adder 202 to output as a final output signal.

[0100] In the above-mentioned configuration, the average noise power output from average noise power calculator 126, LPC output from LSP/LPC converter 212 and scaling coefficient output from scaling calculator 216 both are parameters used in performing the post-processing.

[0101] Thus, according to this embodiment, a noise generated in noise generating section 201 is added to the decoded signal (post-filter output signal), and then scaling section 203 performs the scaling. In this way, since the power of the noise-added decoding signal is subjected to scaling, it is possible to equalize the power of the noise-added decoded signal to the power of the decoded signal to which the noise is not added yet. Further, since the inter-frame smoothing and inter-sample smoothing is both used, the stationary noise becomes smoother, and it is possible to improve the quality of subjective stationary noises.

[0102] (Third Embodiment)

[0103] FIG. 6 illustrates a configuration of a stationary noise post-processing apparatus according to the third embodiment of the present invention. In FIG. 6, the same sections as in FIG. 5 are assigned the same reference numerals as in FIG. 5, and specific descriptions thereof are omitted.

[0104] The apparatus is comprised of the configuration of stationary noise post-processing apparatus 200 as illustrated in FIG. 2, and further provided memories that store parameters required to generating noise signals and scaling when a frame is erased, frame erasure concealment processing control section and switches used in frame erasure concealment processing.

[0105] Stationary noise post-processing apparatus 300 is comprised of noise generating section 301, adder 202, scaling section 303 and frame erasure concealment processing control section 304.

[0106] Noise generating section 301 is comprised of the configuration noise generating section 201 as illustrated in FIG. 5, and further provided memories 310 and 311 that store parameters required to generating noise signals and scaling when a frame is erased, and switches 313 and 314 that are switched on/off in frame erasure concealment processing. Scaling section 303 is comprised of memory 312 that stores parameters required to generating noise signals and scaling when a frame is erased, and switch 315 that is switched on/off in frame erasure concealment processing.

[0107] The operation of stationary noise post-processing apparatus 300 will be described below. First, the operation of noise generating section 301 is explained.

[0108] Memory 310 stores the power (average noise power) of a stationary noise signal output from average noise power calculator 126 via switch 313 to output to gain adjustor 215.

[0109] Switch 313 is switched on/off according to a control signal from frame erasure concealment processing control section 304. Specifically, switch 313 is switched off in the case where the control signal is input which instructs to perform the frame erasure concealment processing, while being switched on in other cases. When switch 313 is switched off, memory 310 stores the power of the stationary noise signal in the last subframe, and outputs the power of the stationary noise signal in the last subframe to gain adjustor 215 when necessary until switch 313 is switched on again.

[0110] Memory 311 stores LPC of the stationary noise signal output from LSP/LPC converter 212 via switch 314 to output to synthesis filter 211.

[0111] Switch 314 is switched on/off according to a control signal from frame erasure concealment processing control section 304. Specifically, switch 314 is switched off in the case where the control signal is input which instructs to perform the frame erasure concealment processing, while being made in other cases. When switch 314 is switched off, memory 311 stores LPC of the stationary noise signal in the last subframe, and outputs LPC of the stationary noise signal in the last subframe to synthesis filter 211 when necessary until switch 314 is switched on again.

[0112] The operation of scaling section 303 will be described below.

[0113] Memory 312 stores a scaling coefficient that is calculated in scaling coefficient calculating section 216 and output via switch 315, and outputs the coefficient to inter-subframe smoother 217.

[0114] Switch 315 is switched on/off according to a control signal from frame erasure concealment processing control section 304. Specifically, switch 315 is switched off in the case where the control signal is input which instructs to perform the frame erasure concealment processing, while being made in other cases. When switch 315 is switched off, memory 312 stores the scaling coefficient in the last subframe, and outputs the scaling coefficient in the last subframe to inter-subframe smoother 217 when necessary until switch 315 is switched on again.

[0115] Frame erasure concealment processing control section 304 receives as its input frame erasure indication obtained by error detection, etc, and outputs the control signal for instructing to perform the frame erasure concealment processing to switches 313 to 315, in a subframe in an erased frame and a subframe (error recovery frame) recovered from an error after an erased frame. There is a case that the frame erasure concealment processing in the error recovery subframe is performed in a plurality of subframes (for example, in two subframes) The frame erasure concealment processing is to prevent the quality of decoded results from deteriorating when information is lost in part of subframes, by using information of a (previous) frame preceding the erased frame. In addition, when extreme power attenuation does not occur at all in the error recovery subframe subsequent to the erasee frame, the frame erasure concealment processing is not required in the error recovery subframe.

[0116] In a generally used frame erasure concealment method, a current frame is extrapolated using previously received information. In this case, since the extrapolated data causes the subjective quality to deteriorate, the signal power is attenuated gently. However, when a frame erasures in a stationary noise region, it happens sometimes that the deterioration of objective quality due to signal discontinuity caused by power attenuation is larger than the deterioration of the subjective equality due to distortion caused by the extrapolation. In particular, in packet communications as typified by internet communications, frames sometimes are erased successively, and the deterioration due to signal discontinuity tends to be remarkable. In order to avoid the quality deterioration caused by the signal discontinuity, in the stationary noise post-processing apparatus according to the present invention, gain adjustor 215 calculates the gain adjustment coefficient to scale up to the average noise power from average power calculator 126 to multiply by the stationary noise signal. Further, scaling coefficient calculator 216 calculates the scaling coefficient to cause the power of the stationary noise signal to which the post-filter output signal is added not to vary greatly, and outputs the signal multiplied by the scaling coefficient as a final output signal. In this way, it is possible to suppress variations in the power of the final output signal to a small level and to maintain the stationary noise signal level obtained before frame erasure, whereby it is possible to suppress deterioration of the subjective quality due to sound signal discontinuity.

[0117] (Fourth Embodiment)

[0118] FIG. 7 is a diagram illustrating a configuration of a speech decoding processing system according to the fourth embodiment of the present invention. The speech decoding processing system is comprised of code receiving apparatus 100, speech decoding apparatus 101 and stationary noise region detecting apparatus 102 that are explained in the first embodiment, and stationary noise post-processing apparatus 300 explained in the third embodiment. In addition, the speech decoding processing system may have stationary noise post-processing apparatus 200 explained in the second embodiment, instead of stationary noise post-processing apparatus 300.

[0119] The operation of the speech decoding processing system will be described below. Specific descriptions of each structural element are stated in the first to third embodiments with reference to FIG. 1, FIG. 5 and FIG. 6, and therefore in FIG. 7, the same sections as in FIG. 1, FIG. 5 and FIG. 6 are assigned the same reference numerals as in FIG. 1, FIG. 5 and FIG. 6 respectively to omit the specific descriptions.

[0120] Code receiving apparatus 100 receives a coded signal from the transmission path, and divides various parameters to output speech decoding apparatus 101. Speech decoding apparatus 101 decodes a speech signal from the various parameters, and outputs a post-filter output signal and required parameters obtained during the decoding processing to stationary noise region detecting apparatus 102 and stationary noise post-processing section 300. Stationary noise region detecting apparatus 102 determines a current subframe is a stationary noise region using the information input form speech decoding apparatus 101, and outputs the determination result and required parameters obtained during the determination processing to stationary noise post-processing apparatus 300.

[0121] With respect to the post-filter output signal input from speech decoding apparatus 101, stationary noise post-processing apparatus 300 performs the processing for generating a stationary noise signal to multiplex on the post-filter output signal, using the various parameter information input from speech decoding apparatus 101 and the determination information and various parameter information input from stationary noise region detecting apparatus 102, and outputs the processing result as a final post-filter output signal.

[0122] FIG. 8 is a flow diagram showing the flow of the processing of the speech decoding system according to this embodiment. FIG. 8 only shows the flow of processing in stationary noise region detecting apparatus 102 and stationary noise post-processing apparatus 300 as illustrated in FIG. 7, and omits the processing in code receiving apparatus 100 and speech decoding apparatus 101, because such processing can be implemented by well-known techniques generally used. The operation of the processing subsequent to speech decoding apparatus 101 in the system will be described below with reference to FIG. 8. First in ST501, various variables stored in memories are initialized in the speech decoding system according to this embodiment. FIG. 9 shows examples of memories to be initialized and initial values.

[0123] Next, the processing of ST502 to ST505 is performed in a loop. The processing is performed until speech decoding apparatus 101 does not output the post-filter output signal (speech decoding apparatus 101 stops the processing). In ST502, mode determination is made, and it is determined whether a current subframe is a stationary noise region (stationary noise mode) or speech region (speech mode). The processing flow in ST502 is explained later specifically.

[0124] In ST503, stationary noise post-processing apparatus 300 performs stationary noise addition (stationary noise post processing). The flow of the stationary noise post processing performed in ST503 is explained later specifically. In ST504, scaling section 303 performs the final scaling processing. The flow of the scaling processing performed in ST504 is explained later specifically.

[0125] In ST505, it is checked whether a subframe is last one to determine whether to finish or continue the loop processing of ST502 to ST505. The loop processing is performed until speech decoding apparatus 101 does not output the post-filter output signal (speech decoding apparatus 101 stops the processing). When the loop processing ends, the processing in the speech decoding system according to this embodiment is all finished.

[0126] The flow of mode determination processing in ST502 will be described below with reference to FIG. 10. First, in ST701, it is checked whether a current subframe is of an erased frame.

[0127] When the current subframe is of an erased frame, the processing flow proceeds to ST702 in which the hangover counter for the frame erasure concealment processing is set for a predetermined value (herein, “3” is assumed), and further proceeds to ST704. The predetermined value for which the hangover counter is set corresponds to the number of frames on which the frame erasure concealment processing is performed continuously even when the subframes are successful (frame erasure does not occur) after the frame erasure occurs.

[0128] When the current subframe is not of an erased frame, the processing flow proceeds to ST703, and it is checked whether a value of the hangover counter for the frame erasure concealment processing is 0. As a result of the check, when the value of the hangover counter for the frame erasure concealment processing is not 0, the value of the hangover counter for the frame erasure concealment processing is decremented by 1, and the processing flow proceeds to ST704.

[0129] In ST704, it is determined whether to perform the frame erasure concealment processing. When the current subframe is neither of an erased frame nor a hangover region immediately after the eraseed frame, it is determined that the frame erasure concealment processing is not performed, and the processing flow proceeds to ST705. When the current subframe is of an erased frame or is a hangover region immediately after the erased frame, it is determined that the frame erasure concealment processing is performed, and the processing flow proceeds to ST707.

[0130] In ST705, the smoothed adaptive code gain is calculated and the pitch history analysis is performed as illustrated in the first embodiment. Since the processing is illustrated in the first embodiment, descriptions thereof are omitted. In addition, the processing flow of the pitch history analysis is explained with reference to FIG. 2. After the processing is performed, the processing flow proceeds to ST706. In ST706, the mode selection is performed. The flow of the mode selection is illustrated specifically in FIGS. 3 and 4. In ST708, the average LSP of the stationary noise region calculated in ST706 is converted into LPC. The processing in ST708 may be not performed subsequent to ST706, and is only required to be performed before a stationary noise signal is generated in ST503.

[0131] When it is determined that the frame erasure concealment processing is performed in ST704, it is set in ST707 that the mode and average LPC of the stationary noise region in the last subframe are used repeatedly respectively as a mode and average LPC in the current subframe, and the processing flow proceeds to ST709.

[0132] In ST709, the mode information (information indicative of whether the current subframe is the stationary noise mode or speech signal mode) in the current subframe and the average LPC of the stationary noise region in the current subframe are stored in the memories. In addition, it is not required to always store the current mode information in the memory in this embodiment, but the current mode information needs to be stored when the mode determination result is used in another block (for example, speech decoding apparatus 101). As described above, the mode determination processing in ST502 is finished.

[0133] The flow of stationary noise addition processing in ST503 will be described below with reference to FIG. 11. First in ST801, excitation generator 210 generates a random vector. Any method of generating a random vector is usable, but the method as illustrated in the second embodiment is effective in which a random vector is selected at random from fixed codebook 113 provided in speech decoding apparatus 101.

[0134] In ST802, using the random vector generated in ST801 as an excitation, LPC synthesis filtering processing is performed. In ST803, the noise signal synthesized in ST802 undergoes the band-limitation filtering processing, so that the bandwidth of the noise signal is adapted to the bandwidth of the decoded signal output from speech decoding apparatus 101. It should be noticed that this processing is not mandatory. In ST804, the power of the synthesized noise signal subjected to band limitation obtained in ST803 is calculated.

[0135] In ST805, the smoothing processing is performed on the signal power obtained in ST804. The smoothing can be implemented readily by performing AR processing as indicated in (Eq.1) in successive frames. The coefficient k of smoothing is determined depending on how much smoothing is required for a stationary signal. It is preferable to perform relatively strong smoothing of about 0.05 to 0.2. Specifically, (Eq.10) is used.

[0136] In ST806, the ratio of the power (already calculated in ST1118) of the stationary noise signal to be generated to the signal power subjected to the inter-subframe smoothing obtained in ST805 is calculated as a gain adjustment coefficient (Eq.11). The calculated gain adjustment coefficient is subjected to the smoothing processing for each sample (Eq.12), and is multiplied by the synthesized noise signal subjected to the band-limitation filtering processing of ST803. The stationary noise signal multiplied by the gain adjustment coefficient is multiplied by a predetermined constant (fixed gain). The fixed gain is multiplied to adjust the absolute level of the stationary noise signal.

[0137] In ST807, the synthesized noise signal generated in ST806 is added to the post-filter output signal output from speech decoding apparatus 101, and the power of the post-filter output signal to which the noise signal is added is calculated.

[0138] In ST808, the ratio of the power of the post-filter output signal output from speech decoding apparatus 101 to the power calculated in ST807 is calculated as a scaling coefficient (Eq.13). The scaling coefficient is used in the scaling processing in ST504 performed downstream of the stationary noise addition processing.

[0139] Finally, adder 202 adds the synthesized noise signal (stationary noise signal) generated in ST806 and the post-filter output signal output from speech decoding apparatus 101. It should be noticed that this processing may be included and performed in ST807. In this way, the stationary noise addition processing in ST503 is finished.

[0140] The flow of scaling in ST504 will be described below with reference to FIG. 12. First in ST901, it is checked whether a current subframe is a target subframe for the frame erasure concealment processing. When the current subframe is a target subframe for the frame erasure concealment processing, the processing flow proceeds to ST902, while proceeding to ST903 when the current subframe is not the target subframe.

[0141] In ST902 the frame erasure concealment processing is performed. In other words, it is set that the scaling coefficient in the last subframe is used repeatedly as a current scaling coefficient, and the processing flow proceeds to ST903.

[0142] In ST903, using the determination result output from stationary noise region detecting apparatus 102, it is checked whether the mode is the stationary noise mode. When the mode is the stationary noise mode, the processing flow proceeds to ST904, while proceeding to ST905 when the mode is not the stationary noise mode.

[0143] In ST904, using (Eq.1) as described previously, the scaling coefficient is subjected to the inter-subframe smoothing processing. In this case, a value of k is set at about 0.1. Specifically, an equation like (Eq.14) is used. The processing is performed to smoothe power variations between subframes in the stationary noise region. After performing the smoothing processing, the processing flow proceeds to ST905.

[0144] In ST905, the scaling coefficient is subjected to smoothing for each sample, and the smoothed scaling coefficient is multiplied by the post-filter output signal to which is added the stationary noise generated in ST502. The smoothing for each sample is also used using (Eq.1), and in this case, a value of k is set at about 0.15. Specifically, an equation like (Eq.15) is used. As described above, the scaling processing in ST504 is finished, thus the scaled post-filter output signal mixed with the stationary noise is obtained.

[0145] In each of the above-mentioned embodiments, equations indicated by (Eq.1) and others are used to calculate the smoothing and average value, but an equation used in smoothing is not limited to such an equation. For example, it may be possible to use an average value in a predetermined previous region.

[0146] The present invention is not limited to the above-mentioned first to fourth embodiments, and is capable of being carried into practice with various modifications thereof. For example, the stationary noise region detecting apparatus of the present invention is applicable to any type of decoder.

[0147] The present invention is not limited to the above-mentioned first to fourth embodiments, and is capable of being carried into practice with various modifications thereof. For example, the above-mentioned embodiments describe cases where the present invention is implemented as a speech decoding apparatus, but are not limited to such cases. The speech decoding method may be performed as software.

[0148] For example, it may be possible that a program for executing the speech decoding method as described above is stored in a ROM (Read Only Memory) in advance, and that the program is executed by a CPU (Central Processor Unit).

[0149] Further, it may be possible to store a program for executing the speech decoding method as described above in a computer readable storage medium, further store the program stored in the storage medium in a RAM (Random Access Memory), and operate a computer according to the program.

[0150] As is apparent from the foregoing, according to the present invention, a degree of periodicity of a decoded signal is determined using an adaptive code gain and pitch periods, and based on the degree of periodicity, it is determined that a subframe is a stationary noise region. Accordingly, it is possible to determine signal states accurately with respect to signals such as sine waves and stationary vowels that are stationary but not noises.

[0151] This application is based on the Japanese Patent Application No.2000-366342 filed on Nov. 30, 2000, entire content of which is expressly incorporated by reference herein.

INDUSTRIAL APPLICABILITY

[0152] The present invention is suitable for use in mobile communication systems, packet communication systems including internet communications and speech decoding apparatuses where speech signals are encoded and transmitted.

Claims

1. A speech decoding apparatus comprising: a first decoding section that decodes a coded signal to obtain at least one type of first parameter indicative of a spectral envelope component of a speech signal; a second decoding section that decodes the coded signal to obtain at least one type of second parameter indicative of a residual component of the speech signal; a synthesis section that constructs a synthesis filter based on the first parameter and that drives the synthesis filter using an excitation signal generated based on the second parameter to generate a decoded signal; a first determining section that determines stationary noise characteristics of the decoded signal based on the first parameter; and a second determining section which determines periodicity of the decoded signal based on the second parameter, and based on a determination result of the periodicity, a determination result of the stationary noise characteristics in the first determining section and the first parameter, further determines whether the decoded signal is a stationary noise region.

2. The speech decoding apparatus according to claim 1, wherein the second parameter includes at least a pitch period, and based on variations in the pitch period between processing units, the second determining section determines the periodicity of the decoded signal.

3. The speech decoding apparatus according to claim 1, wherein the second parameter includes at least an adaptive codebook gain to multiply by an adaptive code vector, and based on the adaptive codebook gain, the second determining section determines the periodicity of the decoded signal.

4. The speech decoding apparatus according to claim 1, further comprising: a variation amount calculating section that calculates a variation amount in spectral envelope parameter between processing units, the first parameter including at least the spectral envelope parameter; and a distance calculating section that calculates a distance between an average value of the spectral envelope parameter in a stationary noise region prior to a current processing unit and the spectral envelope parameter in the current processing unit, wherein the first determining section determines stationary characteristics of the decoded signal generated in the synthesis section, based on the variation amount and the distance, and based on the determination result, further determines the stationary noise characteristics of the decoded signal.

5. The speech decoding apparatus according to claim 4, wherein the variation amount calculating section calculates as the variation amount a square error of the spectral envelope parameter in the current processing unit and the spectral envelope parameter in a last processing unit, the distance calculating section calculates as the distance a square error of the average value of the spectral envelope parameter in the stationary noise region prior to the current processing unit and the spectral envelope parameter in the current processing unit, and the first determining section sets thresholds respectively at least with respect to the square error calculated as the variation amount and the square error calculated as the distance, and when the square error calculated as the variation amount and the square error calculated as the distance are both smaller than set respective thresholds, determines that the decoded signal is stationary.

6. The speech decoding apparatus according to claim 4, further comprising: a pitch history analyzing section which temporarily stores respective pitch periods in a plurality of processing units prior to the current processing unit, groups pitch periods close to each other among the stored pitch periods in the plurality of processing units, and outputs the number of groups in grouping; and a signal power variation calculating section that calculates a variation amount between power of the decoded signal in the current processing unit and the average power of the decoded signal in the stationary noise region prior to the current processing unit, wherein the second determining section determines that the decoded signal is a speech region when the variation amount exceeds a predetermined threshold, determines that the decoded signal is a stationary noise region when the decoded signal is not a speech stationary region, the decoded signal is determined to be stationary in the first determining section and a state in which the variation amount calculated in the variation amount calculating section is less than the predetermined threshold has lasted for a predetermined number of processing units or more, and determines that the decode signal is a speech region when the number of groups output from the pitch history analyzing section is not less than a predetermined threshold or the adaptive code gain is not less than a predetermined threshold.

7. The speech decoding apparatus according to claim 1, further comprising: a post-processing section that multiplies a noise added (mixed) signal by a scaling coefficient to adjust power, the scaling coefficient obtained from the decoded signal generated in the synthesis section and the noise added (mixed) signal obtained by adding (mixing) a pseudo stationary noise signal to (with) the decoded signal.

8. The speech decoding apparatus according to claim 7, further comprising: a scaling section that performs smoothing on the scaling coefficient between processing units only when the second determining section determines that the decoded signal is the stationary noise region.

9. The speech decoding apparatus according to claim 8, further comprising: a storage section that stores at least one type of third parameter used in performing post processing; and a control section that outputs the third parameter in a last processing unit from the storage section when frame erasure occurs in the current processing unit, wherein the post-processing section performs the post processing using the third parameter in the last processing unit.

10. The speech decoding apparatus according to claim 9, wherein the third parameter includes at least the scaling coefficient, and the post-processing section performs the post processing using the scaling coefficient in the last processing unit output from the storage section.

11. The speech decoding apparatus according to claim 7, the post-processing section comprises: a noise generating section that generates a pseudo stationary noise signal; an adding section that adds the decoded signal generated in the synthesis section and the pseudo noise signal to generate a noise added (mixed) decoded signal; and a scaling section that multiplies the scaling coefficient by the noise added (mixed) decoded signal to adjust power.

12. The speech decoding apparatus according to claim 11, wherein the noise generating section comprises: an excitation generating section that selects a random code vector at random from a fixed codebook to generate a noise excitation signal; a second synthesis filter that constructs a second synthesis filter based on a linear predictive coefficient and that drives the second synthesis filter using the noise excitation signal to synthesize a pseudo stationary noise signal; and a gain adjustment section that adjusts gain of the pseudo stationary noise signal synthesized in the second synthesis section.

13. The speech decoding apparatus according to claim 11, wherein the scaling section comprises: a scaling coefficient calculating section that calculates the scaling coefficient based on the decoded signal generated in the synthesis section and the noise added (mixed) decoded signal obtained by adding (mixing) the pseudo stationary noise signal to (with) the decoded signal; a first smoothing section that performs smoothing on the scaling coefficient between processing units; a second smoothing section that performs smoothing on the scaling coefficient on which the first smoothing section performs the smoothing; and a multiplying section that multiplies the scaling coefficient on which the second smoothing section performs the smoothing by the noise added (mixed) decoded signal.

14. A speech decoding method, comprising: decoding at least one type of first parameter indicative of a spectral envelope component of a speech signal; decoding at least one type of second parameter indicative of a residual component of the speech signal; constructing a synthesis filter based on the first parameter, and driving the synthesis filter using an excitation signal generated based on the second parameter to generate a decoded signal; determining stationary noise characteristics of the decoded signal based on the first parameter; and determining periodicity of the decoded signal based on the second parameter, and based on a determination result of the periodicity and a determination result of the stationary noise characteristics, further determining whether the decoded signal is a stationary noise region.

15. A storage medium in which a speech decoding program is stored, the program comprising the procedures of: decoding at least one type of first parameter indicative of a spectral envelope component of a speech signal; decoding at least one type of second parameter indicative of a residual component of the speech signal; constructing a synthesis filter based on the first parameter, and driving the synthesis filter using an excitation signal generated based on the second parameter to generate a decoded signal; determining stationary noise characteristics of the decoded signal based on the first parameter; and determining periodicity of the decoded signal based on the second parameter, and based on a determination result of the periodicity and a determination result of the stationary noise characteristics, further determining whether the decoded signal is a stationary noise region.

16. A speech decoding program to make a computer execute the procedures of: decoding at least one type of first parameter indicative of a spectral envelope component of a speech signal; decoding at least one type of second parameter indicative of a residual component of the speech signal; constructing a synthesis filter based on the first parameter, and driving the synthesis filter using an excitation signal generated based on the second parameter to generate a decoded signal; determining stationary noise characteristics of the decoded signal based on the first parameter; and determining periodicity of the decoded signal based on the second parameter, and based on a determination result of the periodicity and a determination result of the stationary noise characteristics, further determining whether the decoded signal is a stationary noise region.