Echo canceler and echo canceling method

An echo canceler for generating an echo replica signal from a far-end input signal and using the echo replica signal to remove an echo component included in a near-end input signal, thereby obtaining a near-end output signal, the echo canceler comprising:

• a voice detector for detecting an active state and a silent state of the far-end input signal;
• a double-talk detector for detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• a normal adaptive filter section that operates while the far-end input signal is active;
• a silent adaptive filter section for that operates while the far-end input signal is inactive;
• an adaptive mode controller for controlling operations of the normal adaptive filter section and the silent adaptive filter section on the basis of the active state and the silent state of the far-end input signal as detected by the voice detector;
• a background noise calculator for calculating an ambient near-end noise power from the near-end output signal when the silent state is detected by the double-talk detector;
• an echo level detector for detecting an echo power from the near-end output signal when the single-talk state is detected by the double-talk detector;
• an averaging controller for calculating a number of vectors to be averaged from an average value of the ambient near-end noise power calculated by the background noise calculator and an average value of the echo power detected by the echo level detector, wherein
• the silent adaptive filter section includes a data selector for sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a time when the voice detector detects a transition from the silent state to the active state to obtain samples, temporarily storing the samples, and generating data vectors from the stored samples, a vector averager for averaging the number of data vectors designated by the averaging controller to generate an averaged far-end input vector and an averaged near-end input vector, an adaptive filter section for convolving the averaged far-end input vector calculated by the vector averager with an internally stored coefficient vector to generate a simulated echo replica signal, subtracting the simulated echo replica signal from the averaged near-end input vector, updating the internally stored coefficient vector, and supplying the updated coefficient vector to the normal adaptive filter section, and an update controller for halting updating of the internally stored coefficient vector in the silent adaptive filter section in the double-talk state;
• and
• a target value input unit for input and storage of a target echo attenuation value, wherein
• the double-talk detector also calculates an average echo attenuation from a residual signal generated by the normal adaptive filter section and the far-end input signal in the single-talk state, and
• the averaging controller obtains the number of vectors to be averaged from the average echo attenuation calculated by the double-talk detector and the target echo attenuation value.

An echo canceler for generating an echo replica signal from a far-end input signal and using the echo replica signal to remove an echo component included in a near-end input signal, thereby obtaining a near-end output signal, the echo canceler comprising:

• a voice detector for detecting an active state and a silent state of the far-end input signal;
• a double-talk detector for detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• a normal adaptive filter section that operates while the far-end input signal is active;
• a silent adaptive filter section for that operates while the far-end input signal is inactive;
• an adaptive mode controller for controlling operations of the normal adaptive filter section and the silent adaptive filter section on the basis of the active state and the silent state of the far-end input signal as detected by the voice detector;
• a background noise calculator for calculating an ambient near-end noise power from the near-end output signal when the silent state is detected by the double-talk detector;
• an echo level detector for detecting an echo power from the near-end output signal when the single-talk state is detected by the double-talk detector;
• an averaging controller for calculating a number of vectors to be averaged from an average value of the ambient near-end noise power calculated by the background noise calculator and an average value of the echo power detected by the echo level detector, wherein
• the silent adaptive filter section includes a data selector for sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a time when the voice detector detects a transition from the silent state to the active state to obtain samples, temporarily storing the samples, and generating data vectors from the stored samples, a vector averager for averaging the number of data vectors designated by the averaging controller to generate an averaged far-end input vector and an averaged near-end input vector, an adaptive filter section for convolving the averaged far-end input vector calculated by the vector averager with an internally stored coefficient vector to generate a simulated echo replica signal, subtracting the simulated echo replica signal from the averaged near-end input vector, updating the internally stored coefficient vector, and supplying the updated coefficient vector to the normal adaptive filter section, and an update controller for halting updating of the internally stored coefficient vector in the silent adaptive filter section in the double-talk state;
• and
• an attenuation monitoring unit wherein
• the double-talk detector also repeatedly calculates an average echo attenuation from a residual signal generated by the normal adaptive filter section and the far-end input signal in the single-talk state;
• and
• the attenuation monitoring unit calculates a difference between average echo attenuations calculated successively by the double-talk detector;
• and
• the averaging controller obtains the number of vectors to be averaged from the difference calculated by the attenuation monitoring unit.

An echo canceler for generating an echo replica signal from a far-end input signal and using the echo replica signal to remove an echo component included in a near-end input signal, thereby obtaining a near-end output signal, the echo canceler comprising:

• a voice detector for detecting an active state and a silent state of the far-end input signal;
• a double-talk detector for detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• a normal adaptive filter section that operates while the far-end input signal is active;
• a silent adaptive filter section for that operates while the far-end input signal is inactive;
• an adaptive mode controller for controlling operations of the normal adaptive filter section and the silent adaptive filter section on the basis of the active state and the silent state of the far-end input signal as detected by the voice detector;
• a background noise calculator for calculating an ambient near-end noise power from the near-end output signal when the silent state is detected by the double-talk detector;
• an echo level detector for detecting an echo power from the near-end output signal when the single-talk state is detected by the double-talk detector;
• and
• an averaging controller for calculating a number of vectors to be averaged from an average value of the ambient near-end noise power calculated by the background noise calculator and an average value of the echo power detected by the echo level detector,
• wherein the silent adaptive filter section includes a data selector for sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a time when the voice detector detects a transition from the silent state to the active state to obtain samples, temporarily storing the samples, and generating data vectors from the stored samples, a vector averager for averaging the number of data vectors designated by the averaging controller to generate an averaged far-end input vector and an averaged near-end input vector, an adaptive filter section for convolving the averaged far-end input vector calculated by the vector averager with an internally stored coefficient vector to generate a simulated echo replica signal, subtracting the simulated echo replica signal from the averaged near-end input vector, updating the internally stored coefficient vector, and supplying the updated coefficient vector to the normal adaptive filter section, and an update controller for halting updating of the internally stored coefficient vector in the silent adaptive filter section in the double-talk state, and
• wherein the vector averager initially averages all data vectors stored by the data selector, until the number of data vectors stored reaches a predetermined number.

A method of removing an echo of a far-end input signal from a near-end input signal by generating an echo replica from the far-end input signal and subtracting the echo replica signal from the near-end input signal to obtain a near-end output signal, comprising:

• detecting an active state and a silent state of the far-end input signal;
• detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• performing normal adaptive filter operations to generate the echo replica signal while the far-end input signal is active;
• performing silent adaptive filter operations to simulate the normal adaptive filter operations while the far-end input signal is inactive;
• controlling the normal adaptive filter operations and the silent adaptive filter operations according to the active state and the silent state;
• calculating an ambient near-end noise power from the near-end output signal when the silent state is detected;
• detecting an echo power from the near-end output signal when the single-talk state is detected;
• calculating a number of vectors to be averaged from an average value of the ambient near-end noise power and an average value of the echo power;
• wherein the silent adaptive filter operations include
• sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a transition from the silent state to the active state to obtain samples,
• storing the samples temporarily,
• generating data vectors from the samples,
• averaging the calculated number of data vectors to be averaged to generate an averaged far-end input vector and an averaged near-end input vector
• convolving the averaged far-end input vector with a coefficient vector to generate a simulated echo replica signal,
• subtracting the simulated echo replica signal from the averaged near-end input vector to obtain a simulated residual signal,
• updating the coefficient vector according to the simulated residual signal,
• supplying the updated coefficient vector for use in the normal adaptive filter operations, and
• halting updating of the coefficient vector in the double-talk state;
• and
• providing a table of ratios of the average ambient near-end noise power to the average echo power and corresponding predetermined values of the number of vectors to be averaged, wherein calculating the number of vectors to be averaged comprises referring to the table.

A method of removing an echo of a far-end input signal from a near-end input signal by generating an echo replica from the far-end input signal and subtracting the echo replica signal from the near-end input signal to obtain a near-end output signal, comprising:

• detecting an active state and a silent state of the far-end input signal;
• detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• performing normal adaptive filter operations to generate the echo replica signal while the far-end input signal is active;
• performing silent adaptive filter operations to simulate the normal adaptive filter operations while the far-end input signal is inactive;
• controlling the normal adaptive filter operations and the silent adaptive filter operations according to the active state and the silent state;
• calculating an ambient near-end noise power from the near-end output signal when the silent state is detected;
• detecting an echo power from the near-end output signal when the single-talk state is detected;
• calculating a number of vectors to be averaged from an average value of the ambient near-end noise power and an average value of the echo power;
• wherein the silent adaptive filter operations include
• sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a transition from the silent state to the active state to obtain samples,
• storing the samples temporarily,
• generating data vectors from the samples,
• averaging the calculated number of data vectors to be averaged to generate an averaged far-end input vector and an averaged near-end input vector
• convolving the averaged far-end input vector with a coefficient vector to generate a simulated echo replica signal,
• subtracting the simulated echo replica signal from the averaged near-end input vector to obtain a simulated residual signal,
• updating the coefficient vector according to the simulated residual signal,
• supplying the updated coefficient vector for use in the normal adaptive filter operations, and
• halting updating of the coefficient vector in the double-talk state;
• and
• entering a target echo attenuation value;
• and
• calculating an average echo attenuation from a residual signal generated by the normal adaptive filter operations and the far-end input signal in the single-talk state;
• wherein the number of vectors to be averaged is calculated from the average echo attenuation and the target echo attenuation value.

A method of removing an echo of a far-end input signal from a near-end input signal by generating an echo replica from the far-end input signal and subtracting the echo replica signal from the near-end input signal to obtain a near-end output signal, comprising:

• detecting an active state and a silent state of the far-end input signal;
• detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• performing normal adaptive filter operations to generate the echo replica signal while the far-end input signal is active;
• performing silent adaptive filter operations to simulate the normal adaptive filter operations while the far-end input signal is inactive;
• controlling the normal adaptive filter operations and the silent adaptive filter operations according to the active state and the silent state;
• calculating an ambient near-end noise power from the near-end output signal when the silent state is detected;
• detecting an echo power from the near-end output signal when the single-talk state is detected;
• calculating a number of vectors to be averaged from an average value of the ambient near-end noise power and an average value of the echo power;
• wherein the silent adaptive filter operations include
• sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a transition from the silent state to the active state to obtain samples,
• storing the samples temporarily,
• generating data vectors from the samples,
• averaging the calculated number of data vectors to be averaged to generate an averaged far-end input vector and an averaged near-end input vector
• convolving the averaged far-end input vector with a coefficient vector to generate a simulated echo replica signal,
• subtracting the simulated echo replica signal from the averaged near-end input vector to obtain a simulated residual signal,
• updating the coefficient vector according to the simulated residual signal,
• supplying the updated coefficient vector for use in the normal adaptive filter operations, and
• halting updating of the coefficient vector in the double-talk state;
• repeatedly calculating an average echo attenuation from a residual signal generated by the normal adaptive filter operations and the far-end input signal in the single-talk state;
• and
• calculating a difference between successively calculated average echo attenuations;
• wherein the number of vectors to be averaged is calculated from said difference.

A non-transitory machine-readable tangible medium storing a program executable by a computing device to remove an echo of a far-end input signal from a near-end input signal by generating an echo replica from the far-end input signal and subtracting the echo replica signal from the near-end input signal to obtain a near-end output signal, the program comprising instructions for:

• detecting an active state and a silent state of the far-end input signal;
• detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• performing normal adaptive filter operations to generate the echo replica signal while the far-end input signal is active;
• performing silent adaptive filter operations to simulate the normal adaptive filter operations while the far-end input signal is inactive;
• controlling the normal adaptive filter operations and the silent adaptive filter operations according to the active state and the silent state;
• calculating an ambient near-end noise power from the near-end output signal when the silent state is detected;
• detecting an echo power from the near-end output signal when the single-talk state is detected;
• calculating a number of vectors to be averaged from an average value of the ambient near-end noise power and an average value of the echo power;
• wherein the silent adaptive filter operations include
• sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a transition from the silent state to the active state to obtain samples,
• storing the samples temporarily,
• generating data vectors from the samples,
• averaging the calculated number of data vectors to be averaged to generate an averaged far-end input vector and an averaged near-end input vector,
• convolving the averaged far-end input vector with a coefficient vector to generate a simulated echo replica signal,
• subtracting the simulated echo replica signal from the averaged near-end input vector to obtain a simulated residual signal,
• updating the coefficient vector according to the simulated residual signal,
• supplying the updated coefficient vector for use in the normal adaptive filter operations, and
• halting updating of the coefficient vector in the double-talk state;
• and
• providing a table of ratios of the average ambient near-end noise power to the average echo power and corresponding predetermined values of the number of vectors to be averaged, wherein calculating the number of vectors to be averaged comprises referring to the table.

A non-transitory machine-readable tangible medium storing a program executable by a computing device to remove an echo of a far-end input signal from a near-end input signal by generating an echo replica from the far-end input signal and subtracting the echo replica signal from the near-end input signal to obtain a near-end output signal, the program comprising instructions for:

• detecting an active state and a silent state of the far-end input signal;
• detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• performing normal adaptive filter operations to generate the echo replica signal while the far-end input signal is active;
• performing silent adaptive filter operations to simulate the normal adaptive filter operations while the far-end input signal is inactive;
• controlling the normal adaptive filter operations and the silent adaptive filter operations according to the active state and the silent state;
• calculating an ambient near-end noise power from the near-end output signal when the silent state is detected;
• detecting an echo power from the near-end output signal when the single-talk state is detected;
• calculating a number of vectors to be averaged from an average value of the ambient near-end noise power and an average value of the echo power;
• wherein the silent adaptive filter operations include
• sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a transition from the silent state to the active state to obtain samples,
• storing the samples temporarily,
• generating data vectors from the samples,
• averaging the calculated number of data vectors to be averaged to generate an averaged far-end input vector and an averaged near-end input vector,
• convolving the averaged far-end input vector with a coefficient vector to generate a simulated echo replica signal,
• subtracting the simulated echo replica signal from the averaged near-end input vector to obtain a simulated residual signal,
• updating the coefficient vector according to the simulated residual signal, and
• supplying the updated coefficient vector for use in the normal adaptive filter operations;
• halting updating of the coefficient vector in the double-talk state;
• entering a target echo attenuation value;
• and
• calculating an average echo attenuation from a residual signal generated by the normal adaptive filter operations and the far-end input signal in the single-talk state;
• wherein the number of vectors to be averaged is calculated from the average echo attenuation and the target echo attenuation value.

A non-transitory machine-readable tangible medium storing a program executable by a computing device to remove an echo of a far-end input signal from a near-end input signal by generating an echo replica from the far-end input signal and subtracting the echo replica signal from the near-end input signal to obtain a near-end output signal, the program comprising instructions for:

• detecting an active state and a silent state of the far-end input signal;
• detecting a single-talk state, a double-talk state, and a silent state from the far-end input signal and the near-end output signal;
• performing normal adaptive filter operations to generate the echo replica signal while the far-end input signal is active;
• performing silent adaptive filter operations to simulate the normal adaptive filter operations while the far-end input signal is inactive;
• controlling the normal adaptive filter operations and the silent adaptive filter operations according to the active state and the silent state;
• calculating an ambient near-end noise power from the near-end output signal when the silent state is detected;
• detecting an echo power from the near-end output signal when the single-talk state is detected;
• calculating a number of vectors to be averaged from an average value of the ambient near-end noise power and an average value of the echo power;
• wherein the silent adaptive filter operations include
• sampling the near-end input signal and the far-end input signal at prescribed intervals for a prescribed duration starting from a transition from the silent state to the active state to obtain samples,
• storing the samples temporarily,
• generating data vectors from the samples,
• averaging the calculated number of data vectors to be averaged to generate an averaged far-end input vector and an averaged near-end input vector,
• convolving the averaged far-end input vector with a coefficient vector to generate a simulated echo replica signal,
• subtracting the simulated echo replica signal from the averaged near-end input vector to obtain a simulated residual signal,
• updating the coefficient vector according to the simulated residual signal, and
• supplying the updated coefficient vector for use in the normal adaptive filter operations;
• halting updating of the coefficient vector in the double-talk state;
• repeatedly calculating an average echo attenuation from a residual signal generated by the normal adaptive filter operations and the far-end input signal in the single-talk state;
• and
• calculating a difference between successively calculated average echo attenuations;
• wherein the number of vectors to be averaged is calculated from said difference.