Method and system to modify speech impaired messages utilizing neural network audio filters

1. A method to be implemented by a system including memory configured to store program instructions and a neural network (NN) audio filter that includes a connectionist temporal classification (CTC), the system further including one or more processors configured to execute the program instructions, the method comprising:

obtaining, at the system, an electronic audio signal comprising a speech impaired message;

under direction of the one or more processors executing the program instructions, applying the electronic audio signal to the NN audio filter for the CTC to identify an impaired section of the electronic audio signal and modify the electronic audio signal by removing the impaired section thereof to form an unimpaired message; and

outputting the unimpaired message.

2. The method of claim 1 , wherein the impaired section comprises sounds or words that are at least one of distorted or repeated.

3. The method of claim 2 , further comprising identifying, as the impaired section, a sequence of repetitive sounds or words, the sequence of repetitive sounds or words removed to form the unimpaired message.

4. The method of claim 1 , wherein the applying the NN audio filter further comprises: segmenting the electronic audio signal into segments corresponding to time steps; grouping the segments into first and second labels across corresponding first and second collections of the time steps; comparing the first and second labels to identify repetition; and in connection with identifying repetition, removing the segments corresponding to one of the first and second labels.

5. The method of claim 4 , wherein the obtaining includes recording the speech impaired message at a microphone of an electronic device, and wherein the NN audio filter performs the segmenting, grouping, comparing and removing operations in real time while recording the speech impaired message.

6. The method of claim 1 , wherein the NN audio filter comprises a convolutional neural network (CNN) that communicates with a recurrent neural network (RNN), and the RNN communicates with the CTC.

7. The method of claim 1 , further comprising utilizing feature detectors to extract one or more features of interest (FOIs) from segments of the electronic audio signal, identifying word sub-units from the segments, grouping the word sub-units into labels, comparing the labels to identify the impaired section, and removing the word sub-units corresponding to the impaired section to form the unimpaired message.

8. A system, comprising:

a user interface;

memory configured to store program instructions and a neural network (NN) audio filter that includes a connectionist temporal classification (CTC); and

one or more processors that, when executing the program instructions, are configured to:

obtain, at the electronic device, an electronic audio signal comprising a speech impaired message;

apply the electronic audio signal to the NN audio filter for the CTC to identify an impaired section of the electronic audio signal and modify the electronic audio signal by removing the impaired section thereof to form an unimpaired message; and

output the unimpaired message.

9. The system of claim 8 , wherein the one or more processors are configured to identify the impaired section as comprising sounds or words that are at least one of distorted or repeated.

10. The system of claim 9 , wherein the one or more processors are further configured to identify, as the impaired section, a sequence of repetitive sounds or words and remove the sequence of repetitive sounds or words to form the unimpaired message.

11. The system of claim 8 , wherein the one or more processors are further configured to apply the NN audio filter by: segmenting the electronic audio signal into segments corresponding to time steps; grouping the segments into first and second labels across corresponding first and second collections of the time steps; comparing the first and second labels to identify repetition; and in connection with identifying repetition, removing the segments corresponding to one of the first and second labels.

12. The system of claim 11 , further comprising a portable handheld device that includes the user interface, memory, one or more processors and a microphone, the microphone configured to record the speech impaired message, the one or more processors of the handheld device configured to apply the NN audio filter to perform the segmenting, grouping, comparing and removing operations in real time while the microphone records the speech impaired message.

13. The system of claim 8 , wherein the one or more processors, when applying the electronic audio signal to the NN audio filter, are further configured to:

utilize feature detectors to extract one or more features of interest (FOIs) from segments of the electronic audio signal;

identify word sub-units from the segments;

group the word sub-units into labels;

compare the labels to identify the impaired section; and

remove the word sub-units corresponding to the impaired section to form the unimpaired message.

14. The system of claim 8 , wherein the one or more processors are further configured to output the unimpaired message by at least one of: i) replaying the unimpaired message over an audio speaker, ii) transmitting the unimpaired message wired or wirelessly over a network, iii) presenting the unimpaired message on a display as a text message, iv) recording the unimpaired message in the memory on a local electronic device or a remote resource.

15. The system of claim 8 , wherein the NN audio filter comprises a convolutional neural network (CNN) that communicates with a recurrent neural network (RNN), and the RNN communicates with the CTC.

16. The system of claim 15 , wherein the CNN receives the electronic audio signal and extracts a sequence of one or more features of interest (FOIs) from segments of the electronic audio signal, the RNN configured to receive the sequence of one or more FOIs and output value scores for each corresponding element of the sequence of one or more FOIs, the value scores presented in a matrix, the CTC configured to decode the matrix to identify sounds or words that are at least one of distorted or repeated, and identify the impaired section to include the sounds or words that are at least one of distorted or repeated.

17. A computer program product comprising a non-signal computer readable storage medium comprising computer executable code to perform:

obtaining an electronic audio signal comprising a speech impaired message;

applying the electronic audio signal to a neural network (NN) audio filter that includes a connectionist temporal classification (CTC), the computer executable code applies the NN audio filter by utilizing the CTC to identify an impaired section of the electronic audio signal and modify the electronic audio signal by removing the impaired section thereof to form an unimpaired message; and

outputting the unimpaired message.

18. The computer program product of claim 17 , wherein the computer executable code applies the NN audio filter by

identifying the impaired section as comprising sounds or words that are at least one of distorted or repeated.

19. The computer program product of claim 17 , wherein the computer executable code applies the NN audio filter by:

segmenting the electronic audio signal into segments corresponding to time steps;

grouping the segments into first and second labels across corresponding first and second collections of the time steps;

comparing the first and second labels to identify repetition; and

in connection with identifying repetition, removing the segments corresponding to one of the first and second labels.

20. The computer program product of claim 17 , wherein the NN audio filter comprises a convolutional neural network (CNN) that communicates with a recurrent neural network (RNN), and the RNN communicates with the CTC.