Patent Application
20230039619
The present application claims priority from Australian provisional patent application No. 2019904755 filed 16 Dec. 2019, the disclosure of which is hereby incorporated herein by reference.
The present invention relates to a method and apparatus for processing subject sounds for automatic detection of cough sounds therein.
Any references to methods, apparatus or documents of the prior art are not to be taken as constituting any evidence or admission that they formed, or form part of the common general knowledge.
It is known to electronically process subject sounds to predict the presence of respiratory maladies. Where a symptom of the malady is coughing in the subject then it is important to be able to identify segments of the subject sounds that contain coughs, as opposed to background noise for example.
A number of approaches to identifying cough segments of patient sounds are known in the prior art. For example, in WO2013/142908 by Abeyratne at al. there is described a method for cough detection which involves determining a number of features for each of a plurality of segments of a subject's sound, forming a feature vector from those features and applying them to a pre-trained classifier. The output from the classifier is then processed to deem the segments as either “cough” or “non-cough”.
A more recent approach to identifying portions of subject sounds that contain coughs is described in WO 2018/141013 (sometimes called the “LW2” method herein) in which feature vectors from the subject sound are applied to two pre-trained neural nets being respectively trained for detecting an initial phase of a cough sound and a subsequent phase of a cough sound. The first neural net is weighted in accordance with positive training to detect the initial, explosive phase, and the second neural net is positively weighted to detect one or more post-explosive phases of the cough sound. In a preferred embodiment of the LW2 method the first neural net is further weighted in accordance with positive training in respect of the explosive phase and negative training in respect of the post-explosive phases. LW2 is particularly good at identifying cough sounds in a series of connected coughs.
The Inventors have noticed that a problem that can occur with prior art cough identification methods is that they may have undesirably low specificity which means that they identify sound segments as being cough sounds when in fact they are not. Such false positive detection may make those methods infeasible for long term use in high background noise environments where the number of non-cough events in the subject sound recording is much greater than the number of cough events.
It would be desirable if a method and apparatus were provided that can reduce the number of false positives.
A method for identifying cough sounds in an audio recording of a subject comprising:
In an embodiment the method includes, operating said processor to transform the one or more sounds into the image representations wherein the image representations relate frequency and time.
In an embodiment, the one or more image representations comprise spectrograms.
In an embodiment, the one or more image representations comprise mel-spectrograms.
In an embodiment, the method includes, operating said processor to identify the potential cough sounds as cough audio segments of the audio recording by using first and second cough sound pattern classifiers trained to respectively detect initial and subsequent phases of cough sounds.
In an embodiment, the one or more image representations have a dimension of N×M pixels and are formed by said processor processing N windows of each of the cough audio segments wherein each of the N windows is analyzed in M frequency bins.
In an embodiment, each of the N windows overlaps with at least one other of the N windows.
In an embodiment, length of the windows is proportional to length of its associated cough audio segment.
In an embodiment the method includes operating said processor to calculate a Fast Fourier Transform (FFT) and a power value per frequency bin to arrive at a corresponding pixel value of the corresponding image representation of the or more image representations.
In an embodiment the method includes, operating said processor to calculate a power value per frequency bin in the form of M power values, being power values for each of the M frequency bins.
In an embodiment, the M frequency bins comprise M mel-frequency bins, the method including operating said processor to concatenate and normalize the M power values to thereby produce the corresponding image representation in the form of a mel-spectrogram image.
In an embodiment, the image representations are square and wherein M equals N.
In an embodiment, the representation pattern classifier comprises a neural network.
In an embodiment, the neural network is a convolutional neural network (CNN).
In an embodiment the method includes, operating said processor to compare a probability value comprising, or based upon, an output of the representation pattern classifier with a predetermined threshold value.
In an embodiment the method includes, operating said processor to flag one or more of the potential cough sounds as confirmed cough sounds upon the probability value exceeding the predetermined threshold value.
In an embodiment the method includes, operating said processor to flag the confirmed cough sounds by recording begin and end times of the corresponding cough audio segment as being begin and end times of a confirmed cough sound.
In an embodiment the method includes, operating said processor to generate a screen on a display responsive to said processor, the screen indicating the number of potential cough sounds processed and the number of confirmed cough sounds.
According to a further apparatus there is provided an apparatus for identifying cough sounds in a subject comprising:
In an embodiment the apparatus includes, or more cough sound classifiers trained to identify portions of the digital audio recording to thereby produce the pre-identified potential cough sounds.
In an embodiment, the one or more cough sound classifiers comprise a first cough sound pattern classifier and a second cough sound pattern classifiers trained to respectively detect initial and subsequent phases of cough sounds.
In an embodiment, the first cough sound pattern classifier and the and second cough sound pattern classifier each comprise neural networks.
In an embodiment, the sound segment-to-image representation assembly is arranged to transform the pre-identified potential cough sounds into corresponding image representations comprising spectrograms.
In an embodiment, the sound segment-to-image representation assembly is arranged to transform the pre-identified potential cough sounds into corresponding image representations by calculating a Fast Fourier Transform and a power per bin for M to the pre-identified potential cough sounds.
In an embodiment, the sound segment-to-image representation assembly is arranged to transform the pre-identified potential cough sounds into spectrograms
In an embodiment, the spectrograms comprise mel-spectrograms.
In an embodiment, the apparatus includes at least one electronic processor in communication with the electronic memory, wherein the processor is configured by instructions stored in the electronic memory to implement the sound segment-to-image representation assembly.
In an embodiment, the at least one electronic processor is configured by instructions stored in the electronic memory to implement the representation pattern classifier.
In an embodiment, the at least one electronic processor is configured by instructions stored in the electronic memory to implement the at least one cough sound pattern classifier arranged to identify the potential cough sounds.
According to a further aspect of the present invention there is provided a method for training a pattern classifier to confirm a potential cough sound as a confirmed cough sound from a sound recording of the subject, the method comprising:
According to another aspect there is provided a method for identifying cough sounds in an audio recording of a subject including transforming potential cough sounds in the audio recording into corresponding image representations and then applying the image representations to a pre-trained classifier and based on output from the pre-trained classifier flagging the potential cough sounds as confirmed cough sounds or not.
According to a further aspect there is provided an apparatus for processing potential cough sounds identified in an audio recording of a subject, the apparatus including at least one electronic processor in communication with a digital memory storing instructions to configure said processor to implement the method.
According to another aspect of the present invention there is provided a computer readable media bearing tangible, non-transitory machine readable instructions for one or more processors to implement a method for confirming a potential cough sound to be a confirmed cough sound based on an image representation of the potential cough sound.
Preferred features, embodiments and variations of the invention may be discerned from the following Detailed Description which provides sufficient information for those skilled in the art to perform the invention. The Detailed Description is not to be regarded as limiting the scope of the preceding Summary of the Invention in any way. The Detailed Description will make reference to a number of drawings as follows:
A hardware platform that is configured to implement the method comprises a cough identification machine. The machine may be a desktop computer or a portable computational device such as a smartphone that contains at least one processor in communication with an electronic memory that stores instructions that specifically configure the processor in operation to carry out the steps of the method as will be described. It will be appreciated that it is impossible to carry out the method without the specialized hardware, i.e. either a dedicated machine or a machine that is comprised of specially programmed one or more processors. Alternatively, the machine may be implemented as a dedicated assembly that includes specific circuitry to carry out each of the steps that will be discussed. The circuitry may be largely implemented using a Field Programmable Gate Array (FPGA) configured according to a Hardware Descriptor Language (HDL) or Verilog specification.
The processor 53 is in data communication with a plurality of peripheral assemblies 59 to 73, as indicated in
The cough identification machine 51 is programmed with App 56 so that it is configured to operate as a machine for identifying cough segments in the recording of the subject sound.
As previously discussed, although the cough identification machine 51 that is illustrated in
An embodiment of the procedure that cough identification machine 51 uses to identify cough segments in a recording of subject 52, and which comprises instructions that make up App 56 is illustrated in the flowchart of
Initially clinician 54, or another carer or even subject 39, selects App 56 from an app selection screen generated by OS 58 on LCD touchscreen interface 61. In response to that selection the processor 53 displays a screen such as screen 82 of
At box 10 processor 53 identifies potential cough sounds (PCSs) in the audio sound files 50. In a preferred embodiment of the invention the App 56 includes instructions that configure processor 53 to implement a first cough sound pattern classifier (CSPC 1) 62a and a second cough sound pattern classifier (CSPC 2) 62b, each preferably comprising neural networks trained to respectively detect initial and subsequent phases of cough sounds. Thus, in that preferred embodiment the processor 53 identifies the PCSs using the LW2 method that is described in the previously mentioned international patent application publication WO 2018/141013, the disclosure of which is hereby incorporated herein in its entirety by reference. Other methods for identifying potential cough sounds may alternatively be used at box 10, for example the methods described in the previously mentioned international patent publication WO2013/142908 by Abeyratne at al might also be used.
At box 12 the processor 53 sets a variable Current PCS to the first PCS that has been previously identified, i.e. “pre-identified” at box 10.
At box 14 the processor 53 transforms the pre-identified PCS that is stored in the Current PCS variable to produce a corresponding image representation 76 which it stores in either memory 55 or secondary storage 64.
This image representation may comprise, or be based on, a spectrogram of the Current Cough Sound portion of the digital audio file. Possible image representations include mel-frequency spectrogram (or “mel-spectrogram”), continuous wavelet transform, and derivatives of these representations along the time dimension, also known as delta features. Consequently, the image representations relate frequency, for example on a vertical axis, with time for example on a horizontal axis, over the duration of the PCS.
An example of one particular implementation of box 14 is depicted in
Processor 53 identifies the Potential Cough Sounds 66a and 66b as separate cough audio segments 68a and 68b. Each of the separate cough audio segments 68a and 68b are then divided into N, in the present example N=5, equal length overlapping windows 72a1, . . . ,72a5 and 72b1, . . . ,72b5. For a shorter cough segment, e.g. cough segment 68b which is somewhat shorter than cough segment 68a, the overlapping windows 72b that are used to segment section 68b are proportionally shorter to the overlapping windows 72a that are used to segment section 68a.
Processor 53 then calculates a Fast Fourier Transform (FFT) and a power value per mel-bin, for M=5 bins for each of the N=5 windows, to arrive at corresponding pixel values. Machine readable instructions that configure a processor to perform these operations on the sound wave are included in App 56. Such instructions are publicly available, for example at: https://librosa.github.io/librosa/_modules/librosa/core/spectrum.html (retrieved 11 Dec. 2019).
In the example illustrated in
Processor 53 concatenates and normalizes the values stored in the spectrograms 74a and 74b to produce corresponding Square Mel-Spectrogram images 76a and 76b representing cough sounds 66a and 66b respectively. Each of images 76a and 76b is an 8-bit greyscale M×N image where M=N.
N may be any positive integer value bearing in mind that at some N, depending on the sampling rate of the audio interface 71, the cough image will contain all information present in the original audio, which is desirable. The number of FFT bins may need to be increased to accommodate higher N.
In contrast
The images in
Although it is convenient to use square representations that are N×M pixels derived from N segments, each analyzed for M frequency bins, where N=M, it is also possible to use rectangular representations where N is not equal to M provided that the CNN 63 has been trained using similarly dimensioned rectangular images.
From the discussion of box 14 it will be understood that processor 53, configured by App 56 to perform the procedure of box 14, comprises a sound segment-to-image representation assembly that is arranged to transform sound segments of the recording, previously identified as Potential Cough Sounds, e.g. at box 10, into corresponding image representations.
Returning now to
If p is greater than Threshold at box 20 then at box 22 processor 53 flags that the current PCS is a CCS, for example by recording the corresponding sound segment's begin and end times as being the begin and end times of a confirmed cough sound (CCS).
If the p value is not greater than Threshold then the PCS is not flagged as being a CCS. Control then proceeds to decision box 24. At decision box 24 processor 53 checks if there are any more PCSs to be processed. If there are more PCSs, that were identified at box 10, to be processed then at box 26 the Current PCS variable is set to the next identified PCS and control proceeds to box 14 where the previously described boxes 14 to 22 are repeated. If, at box 24, there are no more PCSs to be processed then control proceeds to box 28 where processor 53 operates a display in the form of LCD Touch Screen Interface 61, which is responsive to processor 53, to display the screen 78 shown in
The main board 134 acts as an interface between microprocessors 135 and secondary memory 147. The secondary memory 147 may comprise one or more optical or magnetic, or solid state, drives. The secondary memory 147 stores instructions for an operating system 139. The main board 134 also communicates with random access memory (RAM) 150 and read only memory (ROM) 143. The ROM 143 typically stores instructions for a startup routine, such as a Basic Input Output System (BIOS) or Unified Extensible Firmware Interface (UEFI) which the microprocessor 135 accesses upon start up and which preps the microprocessor 135 for loading of the operating system 139. For example Microsoft Windows, and Ubuntu Linux Desktop are two examples of such an operating system.
The main board 134 also includes an integrated graphics adapter for driving display 147. The main board 133 will typically include a communications adapter 153, for example a LAN adaptor or a modem or a serial or parallel port, that places the server 133 in data communication with a data network.
An operator 167 of CNN training machine 133 interfaces with it by means of keyboard 149, mouse 121 and display 147.
The operator 167 may operate the operating system 139 to load software product 140. The software product 140 may be provided as tangible, non-transitory, machine readable instructions 159 borne upon a computer readable media such as optical disk 157 for reading by disk drive 152. Alternatively, it might also be downloaded via port 153.
The secondary storage 147 also includes software product 140, being a CNN training software product 140 according to an embodiment of the present invention. The CNN training software product 140 is comprised of instructions for CPUs 135 (or as alternatively and collectively referred to “processor 135”) to implement the method that is illustrated in
Initially at box 192 of
At box 196 the processor 135 represents the non-cough events and the cough events as images in the same manner as has previously been discussed at box 14 of
At box 198 processor 135 transforms each image produced at box 196 to create additional training examples for subsequently training a convolutional neural net (CNN). This data augmentation step at box 198 is preferable because a CNN is a very powerful learner and with a limited number of training images it can memorize the training examples and thus over fit the model. The Inventors have discerned that such a model will not generalize well on previously unseen data. The applied image transformations include, but are not limited to, small random zooming, cropping and contrast variations.
At box 200 the processor 135 trains the CNN 142 on the augmented cough and non-cough images that have been produced at box 198 and the original training labels. Over fitting of the CNN 142 is further reduced by using regularization techniques such as dropout, weight decay and batch normalization.
One example of the process used to produce a CNN 142 is to take a pretrained ResNet model, which is a residual network containing shortcut connections, such as ResNet-18, and use the convolutional layers of the model as a backbone, and replace the final non-convolutional layers with layers that suit the cough identification problem domain. These include fully connected hidden layers, dropout layers and batch normalization layers. Information about ResNet-18 is available at: https://www.mathworks.com/help/deeplearning/ref/resnet18.html (retrieved 2 Dec. 2019), the disclosure of which is incorporated herein by reference. ResNet-18 is a convolutional neural network that is trained on more than a million images from the ImageNet database (http://www.image-net.org). The network is 18 layers deep and can classify images into 1000 object categories, such as keyboard, mouse, pencil, and many animals. As a result, the network has learned rich feature representations for a wide range of images. The network has an image input size of 224-by-224 pixels.
The Inventors have found that it is sufficient to fix the ResNet-18 layers and only train the new non-convolutional layers, however it is also possible to re-train both the ResNet-18 layers and the new non-convolutional layers to achieve a working model. A fixed dropout ratio of 0.5 is preferably used. Adaptive Moment Estimation (ADAM) is preferably used as an adaptive optimizer though other optimizer technique may also be used.
At box 202 the original (non-augmented) cough and non-cough images from box 196 are applied to the CNN 142 which is now trained to respond with probabilities for each.
The trained CNN is then distributed as CNN 63 as part of cough identification App 56 being CNN 63.
To test the performance of the method of
75% of that set was used to train the CNN 142 for Deep Cough ID and the remaining 25% (12225 coughs and 4707 non coughs) was used as a test set.
Using LW2, 12225 coughs (PCS) were identified, while 4,707 non-cough events were false positives (i.e. LW2 said these were coughs whereas further investigation revealed that they were not). When Deep Cough ID was used after LW2, 12223 coughs were identified (ie. 2 coughs were false negatives and incorrectly classified), and 4663 non-cough events were now correctly classified (rejected) and only 44 of these non-cough events were incorrectly classified as coughs.
A summary of the performance of the method of
It will be observed from the above table that embodiments of the present invention result in an accuracy increase of over 25% over the prior art LW2 method that is the subject of international patent publication WO 2018/141013.
To recap, in one aspect a method is provided for identifying cough sounds, such as cough sounds 66a, 66b in an audio recording, such as digital sound file 50, of a subject 52. The method in this aspect involves operating at least one electronic processor 53 to identify potential cough sounds (box 10 of
The electronic processor 53 is operated to apply the one or more image representations 76a, 76b to a representation pattern classifier 63 (
In another aspect an apparatus has been described for identifying cough sounds in a subject. The apparatus includes an audio capture arrangement, for example comprised of microphone 75 (
The apparatus has a sound segment-to-image representation assembly arranged to transform pre-identified potential cough sounds into corresponding image representations. For example, the sound segment-to-image representation assembly may comprise processor 53, configured by App 56 to perform the procedure of box 14 (
The apparatus also includes a representation pattern classifier in communication with the sound segment-to-image representation assembly that is configured to process the image representations to thereby produce a signal indicating a probability of the image representations corresponding to the pre-identified potential cough sounds being a confirmed cough sound. The representation pattern classifier may be in the form of a trained convolutional neural network (CNN) 63, which is trained to confirm whether or not the image representation of the Potential Cough Sound is indeed a cough sound i.e. a Confirmed Cough Sound (CCS).
In compliance with the statute, the invention has been described in language more or less specific to structural or methodical features. The term “comprises” and its variations, such as “comprising” and “comprised of” is used throughout in an inclusive sense and not to the exclusion of any additional features.
It is to be understood that the invention is not limited to specific features shown or described since the means herein described comprises preferred forms of putting the invention into effect. The invention is, therefore, claimed in any of its forms or modifications within the proper scope of the appended claims appropriately interpreted by those skilled in the art.
Throughout the specification and claims (if present), unless the context requires otherwise, the term “substantially” or “about” will be understood to not be limited to the value for the range qualified by the terms.
Any embodiment of the invention is meant to be illustrative only and is not meant to be limiting to the invention. Therefore, it should be appreciated that various other changes and modifications can be made to any embodiment described without departing from the spirit and scope of the invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2019904755 | Dec 2019 | AU | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/AU2020/051383 | 12/16/2020 | WO |
