Patent Grant
8135584
This application is the US National Stage of International Application No. PCT/EP2006/000811, filed Jan. 31, 2006 and claims the benefit thereof, which is incorporated by reference herein in their entirety.
The invention relates to a method and arrangements for coding audio signals. In particular the invention relates to a method and an excitation signal generator for forming an excitation signal to excite an audio synthesis filter as well as an audio signal encoder and an audio signal decoder.
In many modern communication systems and in particular in mobile communication systems only limited transmission bandwidths are available for real-time audio transmissions, such as voice or music transmissions for example. In order to transmit as many audio or voice channels as possible in real time by way of a transmission link with limited bandwidth, as via a radio network for example, provision is therefore frequently made for compressing the audio signals to be transmitted by means of realtime-capable or quasi-realtime-capable audio coding methods.
With such audio coding methods the intention is generally to reduce the quantity of data to be transmitted and therefore the transmission rate as much as possible without having too great an adverse effect on the subjective audible impression or in the case of voice transmissions the comprehensibility.
Effective compression of audio signals is also an important aspect in relation to the storage or archiving of audio signals.
Coding methods, wherein an audio signal to be transmitted is aligned time frame by time frame with an audio signal synthesized by means of an audio synthesis filter by optimizing filter parameters, prove to be particularly effective. Such a procedure is frequently also referred to as analysis by synthesis. The audio synthesis filter is hereby excited by an excitation signal that is preferably likewise to be excited. Filtering is frequently also referred to as formant synthesis. The filter parameters used can for example be what are known as LPC coefficients (LPC: Linear Predictive Coding) and/or parameters which specify a spectral and/or temporal envelope of the audio signal. The optimized filter parameters and parameters specifying the excitation signal are then transmitted time frame by time frame to the receiver, to form a synthetic audio signal there by means of an audio synthesis filter provided on the receiver side, said synthetic audio signal being as similar as possible to the original audio signal in respect of the subjective audible impression.
Such an audio coding method is known from the ITU-T recommendation G.729. A realtime audio signal with a bandwidth of 4 kHz can be reduced to a transmission rate of 8 kbit/s by means of the audio coding method described there. According to the G.729 recommendation the excitation signal is generated by means of a so-called adaptive code book in conjunction with a so-called fixed code book. A plurality of predetermined excitation signal sequences are permanently stored in the fixed code book and can be retrieved using a code book index. In contrast already generated excitation signal sequences are stored in the adaptive code book. A respective sequence of the excitation signal is generated by mixing a sequence from the adaptive code book with a sequence from the fixed code book. To optimize the excitation signal, for each time frame both the fixed and adaptive code books are searched for excitation signal sequences, which allow the best possible alignment of the synthetic audio signal with the audio signal to be transmitted. Information relating to access to the sequences found to be optimal from the fixed and adaptive code books is finally transmitted to the receiver as parameters specifying the excitation signal. At the receiver these parameters are used to reconstruct an excitation signal by means of a fixed and adaptive code book of the receiver.
The search through two code books to be carried out according to the G.729 recommendation for each time frame in real time however requires a significant computation outlay, necessitating complex processors.
It is also currently intended to synthesize an audio signal to be transmitted with a higher bandwidth to improve the audible impression. The expansion G.729EV of the G.729 recommendation currently under discussion attempts to expand the audio bandwidth from 4 kHz to 8 kHz.
Such a bandwidth expansion of the synthesized audio signal can be achieved by constructing a suitable excitation signal of higher bandwidth, for example 8 kHz, from a narrowband excitation signal, for example with a bandwidth of 4 kHz, in order to excite the audio synthesis filter over a broad band. Different procedures for forming such a broadband excitation signal are currently under discussion in this context. According to such discussion the broadband excitation signal can be generated by squaring the narrowband excitation signal in the time domain or by generating an expansion band by displacing or mirroring the frequency spectrum of the narrowband excitation signal. However said procedures distort the spectrum of the excitation signal anharmonically and/or a significant, audible phase error is caused in the spectrum.
The object of the present invention is to specify a method for forming an excitation signal for an audio synthesis filter, which allows a further reduction of the transmission rate and/or an improvement in the audible impression and a reduction of the computation outlay required for audio coding during audio signal transmissions. The object of the invention is also to specify an excitation signal generator for implementing the method and an audio signal encoder and an audio signal decoder.
This object is achieved by a method, an excitation signal generator an audio signal encoder and an audio signal decoder with the features of the claims.
Advantageous embodiments and developments of the invention are set out in the dependent claims.
With the inventive method for forming an excitation signal to excite an audio synthesis filter the excitation signal is formed as a series of sampled excitation values. Already formed sampled excitation values are hereby stored in a temporally continuous manner in an adaptive code book. A noise generator is also provided, which generates random sampled values continuously. A sequence of stored sampled excitation values is selected from the adaptive code book based on a supplied audio basic frequency parameter, which predetermines a time interval between the sequence to be selected and the current time reference. The excitation signal is formed by mixing the selected sequence with a random sequence containing current random sampled values of the noise generator.
Using the noise generator as the source of random sampled values means that it is possible to dispense with a fixed code book for filling the adaptive code book. Accordingly it is not necessary to provide or transmit code book indices for selecting predetermined sampled value sequences stored in a fixed code book. Since such code book indices for a fixed code book take up a significant proportion of the audio data to be transmitted with known methods, it is possible generally to reduce the transmission rate to a significant degree with the invention. The saved transmission bandwidth can be used correspondingly for other purposes or to enhance transmission quality.
The noise generator, which preferably generates an essentially white, spectrally flat noise, allows a noise component contained in audio signals or voice signals generally to be modeled better than by means of a fixed code book, which only contains permanently predetermined sampled value sequences. A harmonic fine structure of the audio or voice signals can in contrast be simulated well by the selection of a sampled value sequence from the adaptive code book as a function of the audio basic frequency parameter.
Since a noise generator can naturally be scaled easily to different frequency ranges, bandwidth expansions can be achieved with little outlay. Also the invention prevents a residual coding error being transmitted to an expansion band during bandwidth expansion.
The invention can be deployed advantageously both during encoding and during decoding of an audio signal. In the case of an audio signal encoder an inventive excitation signal generator can excite an audio synthesis filter, whose output audio signal is compared with a respectively current frame of the audio signal to be transmitted. The comparison of the current frame is preferably carried out for different selections of sequences of earlier sampled excitation values stored in the adaptive code book. The temporal position of the sampled value sequence within the adaptive code book, for which the comparison indicates optimal correspondence, can be expressed by a corresponding audio basic frequency parameter, which can then be transmitted to a receiver. A search through a further fixed code book and the additional transmission of code book indices are not necessary.
In the case of an audio signal decoder a respectively received audio basic frequency parameter can control an inventive excitation signal generator in such a manner that it generates an excitation signal corresponding harmonically to the audio basic frequency parameter, without relying on code book indices to be transmitted in addition. The excitation signal thus generated allows an audio synthesis filter to be excited in order to generate a synthetic audio signal, which very closely resembles the original audio signal in respect of audible impression.
This reduces both the necessary computation outlay at the audio signal encoder and the necessary transmission rate. Correspondingly it is possible generally to achieve a higher transmission quality and therefore an improved audible impression for the same transmission rate.
The audio synthesis filters at the audio signal encoder and/or audio signal decoder can be for example in the form of LPC filters, Wiener FIR filters, filters for forming a temporal or spectral envelope of the audio signal or a combination of said filters.
The inventive method can preferably be executed by a signal processor.
According to one advantageous embodiment of the invention the sampled excitation values and/or the random sampled values can be processed time frame by time frame, with the length of the selected sequence and/or the length of the random sequence corresponding to a predetermined length of a time frame.
According to one advantageous development of the invention, if the audio basic frequency parameter predetermines a time interval, which is not a whole-number multiple of a predetermined sampling interval of a narrowband excitation signal to be generated separately, provision can be made to insert intermediate sampled values between the sampled excitation values and/or between the random sampled values as a function of the audio basic frequency parameter. Insertion preferably takes place in such a manner that a sampling interval of the resulting sampled values is smaller than the sampling interval of the narrowband excitation signal. It is thus possible to generate an excitation signal, which has additional frequency components of an expansion band, e.g. from 4-8 kHz compared with a narrowband excitation signal, for example in the frequency range from 0-4 kHz. The excitation signal thus generated has no significant anharmonic distortion other than excitation signals generated by known bandwidth expansion methods.
According to a further embodiment of the invention the selected sequence can be amplified according to a first intensity parameter and/or the random sequence can be amplified according to a second intensity parameter during mixing. The first and second intensity parameters, as well as the audio basic frequency parameter, can preferably be derived from the audio signal to be transmitted and then be transmitted time frame by time frame.
The excitation signal can also be formed with a sampling interval that is smaller compared with a narrowband excitation signal to be generated separately, with the result that the excitation signal has additional frequency components of an expansion band compared with the narrowband excitation signal. In this instance the audio basic frequency parameter and the first and/or second intensity parameter can be derived from audio synthesis parameters, which are actually provided to generate the narrowband excitation signal. Similarly the audio basic frequency parameter and the first and/or second intensity parameter can be derived from a narrowband component of an audio signal to be transmitted.
The audio basic frequency parameter and the first and/or second intensity parameter can therefore be derived from narrowband audio parameters but can be applied to the expansion band. This is advantageous in that no additional audio synthesis parameters are required for bandwidth expansion of the excitation signal outside the audio synthesis parameters provided to generate the narrowband excitation signal. The audio synthesis parameters provided to generate the narrowband excitation signal can generally be provided by existing, narrowband audiocodecs, for example according to the G.729 recommendation.
With known narrowband transmission methods, such as according to the G.729 recommendation for example, the audio basic frequency parameter is frequently determined more accurately than corresponds to the sampling interval of the narrowband excitation signal. An accuracy of for example half or a third of a sampling interval is frequently provided. The audio basic frequency parameter provided for the narrowband excitation signal can thus generally be used directly or essentially unchanged to generate the bandwidth-expanded excitation signal.
The first and/or second intensity parameter can be derived respectively from the corresponding narrowband intensity parameters by applying a predetermined function, in order for example to emphasize a noise component rather than a harmonic component in the expansion band of an audio signal.
A component of the excitation signal ascribable to the expansion band can preferably be combined with the separately generated, narrowband excitation signal, in order to generate a broadband excitation signal, for example in the frequency range from 0 to 8 kHz, to excite the audio synthesis filter.
An exemplary embodiment of the invention is described in more detail below with reference to the drawings containing schematic illustrations, in which:
a to 2b show different embodiments of an inventive excitation signal generator,
The upper partial figure shows the audio signal sampled at a sampling rate of 8 kHz for example. The sampling rate of 8 kHz corresponds to a sampling interval DT1 of 1/8000 s. Audio signals essentially up to a frequency of 4 kHz can be shown by the sampled values sampled at a sampling rate of 8 kHz according to a fundamental sampling theorem. This frequency range is hereafter referred to as narrowband.
The lower partial figure illustrates the audio signal sampled at a sampling rate of 16 kHz. In accordance with the sampling rate, which is double the sampling rate of the upper partial figure, the sampling interval DT2 in the lower partial figure is half the sampling interval DT1, in other words 1/16000 s here. An audio signal essentially up to a frequency of 8 kHz can be shown by the sampled values sampled at a sampling rate of 16 kHz. The above frequency range is also referred to below as broadband. It is obvious that the terms narrowband and broadband are not limited to the frequency ranges simply given by way of example but can be applied generally to any frequency ranges, in so far as the term broadband is intended to specify a larger frequency range than the term narrowband.
a and 2b show schematic diagrams of different embodiments of an inventive excitation signal generator. The excitation signal generators shown respectively comprise a noise generator NOISE, an adaptive code book ACB and a mixing facility MIX as function components. The random generator NOISE serves to generate random sampled values with a respectively predetermined sampling interval in a temporally continuous manner. It should be assumed by way of example for both embodiments shown in
A noise signal formed by the random sampled values preferably has an essentially white or flat frequency spectrum.
Let us look first in the following at the embodiment of the excitation signal generator shown in
While the noise generator NOISE generates random sequences EXC_N, i.e. time frames with random sampled values, continuously, the adaptive code book ACB outputs sequences, i.e. time frames EXC_P of stored sampled excitation values, continuously. The random sequences EXC_N and the sequences EXC_P output by the adaptive code book ACB are routed to the mixing facility MIX, to which intensity parameters G_N for level control of the random sequences EXC_N and intensity parameters G_P for level control of the sequences EXC_P coming from the adaptive code book ACB are also routed time frame by time frame. In the mixing facility MIX the random sampled values of a respective random sequence EXC_N are multiplied, i.e. amplified, time frame by time frame by a respective intensity parameter G_N and the sampled values of a respective sequence EXC_P output by the adaptive code book ACB are multiplied, i.e. amplified, time frame by time frame by a respective intensity parameter G_P. The multiplications are shown in
With regard to the excitation signal EXC the adaptive code book ACB acts as a shift register, in which currently formed sequences of the excitation signal EXC are stored, with previously formed sequences of the excitation signal being displaced successively backward whilst maintaining the temporal order.
The output of the sequences EXC_P of stored sampled excitation values is controlled by audio basic frequency parameters PITCH supplied time frame by time frame to the adaptive code book ACB. The audio basic frequency parameters PITCH are used to select the sequences EXP to be output by the adaptive code book ACB from the stored sampled excitation values. The selection is made by a selection facility SEL of the adaptive code book ACB. Such an audio basic frequency parameter PITCH is frequently also referred to in technical circles as pitch lag.
It should be assumed below that the audio basic frequency parameters PITCH are respectively predetermined in units of a narrowband sampling interval, here for example 1/8000 s with a narrowband sampling rate of 8 kHz. A period of a basic frequency of the audio signal to be transmitted or to be synthesized is specified respectively time frame by time frame by the audio basic frequency parameters PITCH. With modern audio coding methods, e.g. according to the G.729 recommendation, the basic frequency periods of an audio signal are frequently measured or provided with higher resolution than corresponds to a respectively used sampling interval. Such an audio basic frequency parameter, accurate to fractions of sampling intervals, can thus also have values that are not whole numbers in units of the sampling interval. Such an audio basic frequency parameter PITCH, which is not a whole number, contains information about higher frequency components than actually correspond to the sampling interval. While such higher frequency components are filtered out with known audio coders, for example according to the G.729 recommendation, the information about the higher frequency components can be used in a simple manner to improve audio synthesis quality with inventive audio signal generators.
A current time reference T0 is shown in
The sequence EXC_P of stored sampled excitation values, whose start has a time interval from the current time reference T0 corresponding to the audio basic frequency parameter PITCH and whose length corresponds to the predetermined length of a time frame, is selected from the adaptive code book ACB to be output. The time interval here is calculated temporally backward from the current time reference T0. It should be noted that the start of the selected sequence EXC_P does not have to coincide with a time frame limit but in some instances can coincide within predetermined limits with any stored sampled excitation value.
It is assumed by way of example in
When the inventive excitation signal generator is activated, the adaptive code book ACB is initially empty, then to be filled successively with formed sampled excitation values of the output excitation signal EXC. Since the adaptive code book ACB is empty at first, the excitation signal EXC is initially only supplied by the noise generator NOISE as the single signal source. This means that the adaptive code book ACB is first filled with non-periodic random sampled values. In this scenario the question arises as to how periodic signal components can be obtained by means of the adaptive code book ACB, since only a non-periodic noise generator NOISE is available as the original signal source. In fact it was deemed necessary according to former thinking to provide a fixed code book as well as an adaptive code book, in order to fill the adaptive code book ACB with determined signal sequences stored in the fixed code book.
According to research by the inventor however such a fixed code book is not necessary. In fact it is possible to generate an excitation signal with a harmonic fine structure by continuous appropriate selection of sampled value sequences EXC_P from the adaptive code book ACB even without a fixed code book. To clarify the underlying active principle, we will look at an instance where the audio basic frequency parameter PITCH remains constant over a number of time frames. In such an instance a time frame with the same temporal position relative to T0 is read out from the adaptive code book ACB a number of times in succession, mixed with a random sequence EXC_N of the noise generator NOISE and stored again as the current time frame of the excitation signal EXC in the current code book ACB. The current time frame is hereby stored with an interval specified by the audio basic frequency parameter PITCH in relation to the previously output sequence EXC_P. This causes a periodic signal component to form successively in the adaptive code book ACB, its period being determined by the audio basic frequency parameter PITCH. The periodic component of the overall excitation signal EXC is hereby controlled by the intensity parameters G_N and G_P.
Using the noise generator NOISE instead of a fixed code book means that it is not necessary to transmit code book indices for a fixed code book. This means that the transmission rate or bandwidth for the transmission of audio signals can be reduced significantly. Also using the noise generator NOISE allows a better audible impression to be achieved, particularly when playing back non-harmonic or noise-type audio components.
An embodiment of the inventive excitation signal generator for generating a bandwidth-expanded excitation signal EXC is described below with reference to
Let it be assumed by way of example that the bandwidth expansion factor N has a value of 2 and that with a narrowband sampling rate of 8 kHz the sampling rate of the excitation signal EXC to be output is correspondingly N×8 kHz=16 kHz.
While the noise generator NOISE outputs random sampled values with the narrowband sampling rate of 8 kHz, the adaptive code book ACB and the mixing facility MIX use the broadband sampling rate of 16 kHz. To adjust the narrowband sampling rate of the noise generator NOISE to the broadband sampling rate of the mixing facility MIX, an interpolator INT_N is connected between said mixing facility MIX and the noise generator NOISE. The interpolator INT_N receives the random sampled values output by the noise generator NOISE with a narrowband sampling rate and inserts an intermediate sampled value with amplitude 0 between two of these random sampled values respectively. For other values of the bandwidth expansion factor N, N−1 intermediate sampled values, each with amplitude 0, are inserted similarly between two random sampled values respectively. This converts a narrowband white noise spectrum of the noise generator NOISE to a broadband white spectrum.
Let it be assumed that the audio basic frequency parameter PITCH is supplied in units of the narrowband sampling interval. Let it be further assumed that the audio basic frequency parameter PITCH is provided in these units with an accuracy at least to the nearest fraction 1/N, in other words here to the nearest ½. The non-whole-number audio basic frequency parameter PITCH contains information about frequency components outside the narrowband frequency range. Such a non-whole-number audio basic frequency parameter PITCH is frequently also represented by pitch=p+p_frac/N, where p and p_frac are whole-number parameters with p_frac=0, . . . , N−1. Since the adaptive code book ACB uses a sampling interval that is halved or divided by N compared with the narrowband sampling interval, the audio basic frequency parameter PITCH is first multiplied by N. The resulting product PITCH×N=p×N+p_frac is then used to select the stored sampled value sequence EXC_P, as already explained in relation to
The excitation signal generator shown in
Generation of the broadband excitation signal S_EXC is divided into different layers, namely one layer for the narrowband frequency range, in this instance from 0 to 4 kHz, and one layer for the expansion band, in this instance from 4 to 8 kHz. To generate a narrowband excitation signal N_EXC, in this instance in the frequency range from 0 to 4 kHz, the audio signal decoder has a narrowband excitation signal generator NBC and to generate a frequency-expanded excitation signal E_EXC, in this instance in the frequency range from 4 to 8 kHz, it has an excitation signal generator EBC according to
The audio basic frequency parameter PITCH and the intensity parameters G_N and G_P are supplied respectively to the narrowband excitation signal generator NBC time frame by time frame. A sum parameter G_S+G_N and a ratio parameter G_S/G_N or its core value can also be supplied instead of the intensity parameters G_N and G_P.
The audio basic frequency parameter PITCH is not a whole number as already described in conjunction with
The parameters PITCH, G_S and G_N used by the narrowband excitation signal generator NBC are routed to the excitation signal generator EBC equipped according to
With the inventive audio signal decoder only the audio parameters PITCH, G_S and G_N are required to generate the bandwidth-expanded excitation signal E_EXC and therefore to generate the broadband excitation signal S_EXC and these are transmitted anyway to generate the narrowband excitation signal or are supplied by a narrowband excitation signal generator. The audio parameters PITCH, G_N and G_P can thus advantageously be derived from the narrowband frequency range of the audio signal to be transmitted or from parameters of a narrowband codec, in order then to be applied to an expansion band to be added. To generate the broadband excitation signal S_EXC no additional audio parameters have to be transmitted compared with generation of the narrowband excitation signal N_EXC. Dispensing with a fixed code book in the excitation signal generators EBC and/or NBC means that there is also no need for the additional transmission of code book indices. Additional information about an audio structure in the expansion band can be transmitted by the parameters F_ENV and T_ENV.
The audio signal decoder shown in
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/EP2006/000811 | 1/31/2006 | WO | 00 | 7/29/2008 |
| Publishing Document | Publishing Date | Country | Kind |
|---|---|---|---|
| WO2007/087823 | 8/9/2007 | WO | A |
| Number | Name | Date | Kind |
|---|---|---|---|
| 5623575 | Fette et al. | Apr 1997 | A |
| 6047254 | Ireton et al. | Apr 2000 | A |
| 20090012782 | Geiser et al. | Jan 2009 | A1 |
| Number | Date | Country |
|---|---|---|
| 1297222 | May 2001 | CN |
| 0 883 107 | Dec 1998 | EP |
| 1089258 | Apr 2001 | EP |
| Number | Date | Country | |
|---|---|---|---|
| 20090012782 A1 | Jan 2009 | US |
