Unified deep neural network model for acoustic echo cancellation and residual echo suppression
A method, computer program, and computer system is provided for an all-deep-learning based AEC system by recurrent neural networks. The model consists of two stages, echo estimation stage and echo suppression stage, respectively. Two different schemes for echo estimation are presented herein: linear echo estimation by multi-tap filtering on far-end reference signal and non-linear echo estimation by single-tap masking on microphone signal. A microphone signal waveform and a far-end reference signal waveform are received. An echo signal waveform is estimated based on the microphone signal waveform and a far-end reference signal waveform. A near-end speech signal waveform is output based on subtracting the estimated echo signal waveform from the microphone signal waveform, and echoes are suppressed within the near-end speech signal waveform.
1. A method of acoustic echo suppression, executable by a processor, comprising:
receiving a microphone signal waveform and a far-end reference signal waveform;
estimating an echo signal waveform based on multiplying a single-tap mask by a signal spectrum associated with the microphone signal waveform and a far-end reference signal waveform; and
outputting a near-end speech signal waveform based on subtracting the estimated echo signal waveform from the microphone signal waveform, wherein echoes are suppressed within the near-end speech signal waveform.
2. The method of claim 1 , wherein the echo signal is computed based on linear filtering on the far-end reference signal waveform.
3. The method of claim 1 , wherein the echo signal is estimated by a first stage and the near-end speech signal is estimated by a second stage.
4. The method of claim 3 , wherein the first stage applies a scale-invariant signal to distortion ratio on time domain signals associated with the microphone signal waveform and the far-end reference signal waveform and a loss on a time-frequency spectral magnitude associated with the microphone signal waveform and the far-end reference signal waveform.
5. The method of claim 3 , wherein the second stage applies a scale-invariant signal to distortion ratio on time domain signals associated with the microphone signal waveform and the echo signal waveform and a loss on a time-frequency spectral magnitude associated with the microphone signal waveform and the echo signal waveform.
6. The method of claim 3 , wherein the first stage and the second stage comprise a recurrent neural network.
7. A computer system for acoustic echo suppression, the computer system comprising:
one or more computer-readable non-transitory storage media configured to store computer program code; and
one or more computer processors configured to access said computer program code and operate as instructed by said computer program code, said computer program code including:
receiving code configured to cause the one or more computer processors to receive a microphone signal waveform and a far-end reference signal waveform;
estimating code configured to cause the one or more computer processors to estimate an echo signal waveform based on multiplying a single-tap mask by a signal spectrum associated with the microphone signal waveform and a far-end reference signal waveform; and
outputting estimating code configured to cause the one or more computer processors to output a near-end speech signal waveform based on subtracting the estimated echo signal waveform from the microphone signal waveform, wherein echoes are suppressed within the near-end speech signal waveform.
8. The computer system of claim 7 , wherein the echo signal is computed based on linear filtering on the far-end reference signal waveform.
9. The computer system of claim 7 , wherein the echo signal is estimated by a first stage and the near-end speech signal is estimated by a second stage.
10. The computer system of claim 9 , wherein the first stage applies a scale-invariant signal to distortion ratio on time domain signals associated with the microphone signal waveform and the far-end reference signal waveform and a loss on a time-frequency spectral magnitude associated with the microphone signal waveform and the far-end reference signal waveform.
11. The computer system of claim 9 , wherein the second stage applies a scale-invariant signal to distortion ratio on time domain signals associated with the microphone signal waveform and the echo signal waveform and a loss on a time-frequency spectral magnitude associated with the microphone signal waveform and the echo signal waveform.
12. The computer system of claim 9 , wherein the first stage and the second stage comprise a recurrent neural network.
13. A non-transitory computer readable medium having stored thereon a computer program for acoustic echo suppression, the computer program configured to cause one or more computer processors to:
receive a microphone signal waveform and a far-end reference signal waveform;
estimate an echo signal waveform based on multiplying a single-tap mask by a signal spectrum associated with the microphone signal waveform and a far-end reference signal waveform; and
output a near-end speech signal waveform based on subtracting the estimated echo signal waveform from the microphone signal waveform, wherein echoes are suppressed within the near-end speech signal waveform.
14. The computer readable medium of claim 13 , wherein the echo signal is computed based on linear filtering on the far-end reference signal waveform.
15. The computer readable medium of claim 13 , wherein the echo signal is estimated by a first stage and the near-end speech signal is estimated by a second stage.
16. The computer readable medium of claim 15 , wherein the first stage applies a scale-invariant signal to distortion ratio on time domain signals associated with the microphone signal waveform and the far-end reference signal waveform and a loss on a time-frequency spectral magnitude associated with the microphone signal waveform and the far-end reference signal waveform.
17. The computer readable medium of claim 15 , wherein the second stage applies a scale-invariant signal to distortion ratio on time domain signals associated with the microphone signal waveform and the echo signal waveform and a loss on a time-frequency spectral magnitude associated with the microphone signal waveform and the echo signal waveform.