Method and device for detecting double-talk, and echo canceler

A double talk detecting apparatus applicable to an echo canceler which is employed in a telephone line network for performing a transmission between a speech transmitted coming through a four-wire circuit and another speech coming through a two-wire circuit, comprising:

• first feature extraction means for extracting a first feature from a first speech signal corresponding to a speech on said four-wire circuit side, said first feature being specified by a waveform of said first speech signal;
• second feature extraction means for extracting a second feature from a second speech signal corresponding to a speech on said two-wire circuit side, said second feature being specified by a waveform of said second speech signal;
• double talk judgment means for comparing said first feature with said second feature and judging whether or not a double talk is present, based on a result of the comparison;
• wherein said first feature is a result of sampling of said first speech signal, said second feature is a result of sampling of said second speech signal, and said double talk detection means comprises:
• echo path estimation means for estimating an echo path by a learning identification algorithm;
• and
• storage means for preliminarily storing a state corresponding the result of estimation when the double talk does not present, content stored in said storage means being compared with the result of estimation made by said echo path estimation means and a fact as to whether or not a double talk is present is detected based on the comparison.

An echo canceler to be employed in a telephone line network for performing a transmission between a speech coming through a four-wire circuit and another speech coming through a two-wire circuit, comprising:

• echo path estimation means for estimating an echo path through a learning identification algorithm and outputting a tap coefficient corresponding to the result of estimation;
• an echo replica generator for generating an echo replica through a convolutional operation performed based on said tap coefficient;
• storage means for storing an impulse response characteristic of an echo path;
• judgment means for comparing the tap coefficient output by said echo path estimation means with an impulse response characteristic in said storage means and judging that a double talk is present when the result of comparison exceeds a predetermined allowable value;
• and
• control means for stopping the output of the tap coefficient from said echo path estimation means when said judgment means judged that a double talk is present.

A double talk detecting method applicable to an echo canceler which is employed in a telephone line network for performing a transmission between a speech coming through a four-wire circuit and another speech coming through a two-wire circuit, comprising:

• a first step of extracting a first speech pitch from a first speech signal corresponding to a speech on said four-wire circuit side, said first speech pitch being specified by a waveform of said first speech signal;
• a second step of extracting a second speech pitch from a second speech signal corresponding to a speech on said two-wire circuit side, said second speech pitch being specified by a waveform of said second speech signal;
• a third step of comparing said first speech pitch with said second speech pitch and judging whether or not a double talk is present, based on a result of the comparison;
• wherein said third step further comprises the steps of calculating correlation values of fluctuations of the speech pitches of said first and second speech signals;
• comparing said correlation values with a predetermined threshold;
• and
• judging whether or not a double talk is present, based on a result of the comparison.

A double talk detecting apparatus applicable to an echo canceler which is employed in a telephone line network for performing a transmission between a speech transmitted coming through a four-wire circuit and another speech coming through a two-wire circuit, comprising:

• first feature extraction means for extracting a first feature from a first speech signal corresponding to a speech on said four-wire circuit side, said first feature being specified by a waveform of said first speech signal;
• second feature extraction means for extracting a second feature from a second speech signal corresponding to a speech on said two-wire circuit side, said second feature being specified by a waveform of said second speech signal;
• double talk judgment means for comparing said first feature with said second feature and judging whether or not a double talk is present, based on a result of the comparison;
• wherein said first feature is a result of sampling of said first speech signal, said second feature is a result of sampling of said second speech signal, and said double talk detection means comprises:
• frequency band division means for dividing said first and second speech signals into a plurality of frequency bands, respectively and outputting the same;
• comparator means for comparing said first and second speech signals in each of said frequency bands and obtaining power rations therebetween;
• and
• double talk detection means for judging whether or not a double talk is present based on the power ration in each of said frequency bands.