JP2006343644A - Speech recognition method, speech recognition apparatus, program, and recording medium - Google Patents

Abstract

In a speech recognition method for recognizing separately a speech signal and a reception signal of a telephone, a reduction in speech recognition rate due to a side sound component that wraps around the reception signal is improved.
In a speech recognition apparatus that extracts a transmission signal and a reception signal from a telephone and recognizes the transmission signal and the reception signal separately, a side-tone gain estimation for estimating a gain that the transmission signal wraps around the reception signal. A reception signal maximum amplitude estimation unit that estimates the maximum amplitude of the reception signal, a side sound signal estimation unit that estimates a side sound signal for the transmission signal using the estimated side sound gain and reception signal maximum amplitude, A side-tone suppression processing unit that suppresses the side-tone signal from the received signal by the received side-tone signal, and a voice recognition processing unit that recognizes the received voice and the transmitted voice after the side-tone suppression processing separately.
[Selection] Figure 1

Description

  The present invention relates to a voice recognition method, a voice recognition device, a voice recognition program, and a recording medium on which this program is recorded, for taking out a transmission signal and a reception signal from a telephone and recognizing the transmission signal and the reception signal separately. About.

As shown in FIG. 3, a circuit corresponding to a transmission signal connected to the microphone unit 11 of the handset 18 from a line between the telephone set 19 and the handset 18 and a line corresponding to a reception signal connected to the speaker unit 12 of the handset 18. There is a branching device 20 (for example, Patent Document 1), and the speech recognition device 110 that recognizes the speech signal and the reception signal separately by dividing the transmission terminal and the reception terminal of the branching device 20 easily. Conceivable. It is possible to extract only the transmission signal by the conventional branching device 20.
JP 60-223370 A

  In the normal telephone set 19, the side sound in which the transmission signal wraps around the reception signal in the side sound circuit 15 inside the telephone set 19 is superimposed on the reception signal, and it is difficult to extract only the reception signal. Further, when the level of the transmission signal is relatively high or the gain of the side sound circuit 15 is high, the level of the transmission signal wraps around the reception signal as a side sound signal is high, or the side sound level is the maximum amplitude of the reception signal. In the case where saturation is caused by reaching the level, learning of the adaptive filter used in the echo canceller does not proceed well in the saturation section, and therefore, the gain of the side sound signal cannot be estimated. As a result, it is difficult to suppress the side-tone signal using the echo canceller, and there is a drawback that the voice recognition rate of the received voice is lower than that of the transmitted voice.

  An object of the present invention is to provide a speech recognition method and apparatus capable of improving the speech recognition rate of received speech in a speech recognition method and device for recognizing separately transmitted speech and received speech. is there.

  A voice recognition apparatus according to the present invention extracts a transmission signal and a reception signal from a telephone, and estimates a gain at which the transmission signal wraps around the reception signal in a voice recognition apparatus for recognizing the transmission signal and the reception signal separately. Side-tone gain estimating section, received-signal maximum amplitude estimating section for estimating the maximum amplitude of the received signal, and side-tone signal for estimating the side-tone signal for the transmitted signal using the estimated side-tone gain and received-signal maximum amplitude An estimation unit, a side tone suppression unit that suppresses the side tone signal from the received signal based on the estimated side tone signal, and a voice recognition unit that separately recognizes the received voice and the transmitted voice after the side tone suppression processing It is characterized by that.

  According to the speech recognition apparatus of the present invention, since the side sound gain level is always estimated, the transmitted signal component called the side sound is suppressed from the received signal, so that the transmitted signal and the received signal are separated as much as possible. It becomes possible to recognize. That is, even when the speech recognition process is performed on the received signal, higher recognition performance can be obtained by performing the speech recognition process on the signal from which the side sound component is removed than the signal including the side sound component. In addition, side sound level gain is estimated even for side sound signals that cause saturation, so side sound signal suppression processing is possible, thereby reducing the speech recognition rate of the received signal. Can be suppressed.

The speech recognition method and speech recognition apparatus according to the present invention can be realized by hardware. However, in order to achieve a simpler implementation, the speech recognition program according to the present invention is installed in a computer, and the speech according to the present invention is installed in the computer. The embodiment in which the recognition method is executed is the best embodiment.
In order for a computer to execute the speech recognition method according to the present invention, the computer may be operated according to the following procedure.
Sidetone gain estimation process for estimating the gain that the transmitted signal wraps around the received signal, received signal maximum amplitude estimation process for estimating the maximum amplitude of the received signal, and transmission using the estimated sidetone gain and received signal maximum amplitude Side sound signal estimation process for estimating the side sound signal for the signal, side sound suppression process for suppressing the side sound signal from the received signal by the estimated side sound signal, and the received voice and the transmitted voice after the side sound suppression processing. A voice recognition process is performed separately for voice recognition to function as a voice recognition device.

FIG. 1 shows an embodiment of a speech recognition apparatus according to the present invention. Parts corresponding to those in FIG. 3 are denoted by the same reference numerals. In this embodiment, the transmission signal and the reception signal are separately extracted from the branching device 20 and input to the voice recognition device 110 as in the case of FIG.
In the present invention, the speech recognition device 110 is provided with a received signal maximum amplitude estimating unit 21 for estimating a received signal maximum amplitude from a received signal taken out from the branching device 20, a side sound gain estimating unit 22, and a side sound suppression processing unit 23. It is done.

  The received signal maximum amplitude estimating unit 21 estimates the maximum amplitude level from a data series in which the received signal is accumulated for a certain time (for example, about 1 second or more). The side tone gain estimation unit 22 detects a non-maximum amplitude level section that does not reach the maximum amplitude level by using the maximum amplitude level estimated by the reception signal maximum amplitude estimation unit 21 as a clue, and the transmission signal is detected in the non-maximum amplitude level section. Sidetone gain is estimated from the relationship of the received signal.

The side tone suppression processing unit 23 estimates a side tone signal for the transmitted signal from the side tone gain estimated by the side tone gain estimating unit 22 and the maximum amplitude level, and performs a side tone signal suppression process on the received signal. In this way, a reception signal that does not include a side tone signal is obtained.
The time delay between the side sound signal and the transmission signal is relatively small, and a certain degree of suppression performance can be obtained even if the time delay is ignored. By updating the maximum amplitude level and sidetone gain for each frame, the accuracy is improved according to the signal length.

  If the received signal is superimposed on the section where the side sound signal has been saturated, there is a possibility that a large distortion will occur in the suppression process. The signal is low in reliability and difficult to return to a clean signal, and speech recognition performance cannot be expected. However, in the present invention, since a section where saturation is caused by a side sound caused by a transmission signal is known, for example, the signal is padded with zeros. Therefore, it is possible to perform suppression processing that hardly causes adverse effects on voice recognition.

FIG. 2 shows a flowchart for explaining the outline of the speech recognition program according to the present invention.
The received signal maximum amplitude estimating unit 21 estimates the maximum amplitude level from the received signal (steps S1 and S2).
The side-tone gain estimation unit 22 calculates and estimates a side-tone gain (received / send-signal ratio) from the transmitted / received signal in the received signal interval lower than the received signal maximum amplitude level (steps S3 and S4).
After the side sound suppression processing unit 23 multiplies the transmission signal by the estimated side sound gain, the side sound signal is estimated to have an amplitude with the received signal maximum amplitude level as an upper limit (step S5).

The side tone signal is suppressed by subtracting the side tone signal from the received signal (step S6). Finally, recognition processing is performed on the signal after suppression processing (step S7).
As described above, according to the present invention, by suppressing the side sound signal from the reception signal, it is possible to recognize the voice by dividing the transmission signal and the reception signal as much as possible. Therefore, even when the speech recognition process is performed on the received signal, higher recognition performance can be obtained by performing the speech recognition process on the signal from which the side sound component is removed than the signal including the side sound signal. Further, since the side sound level gain is estimated even for the side sound signal that causes saturation, suppression processing can be performed.

The speech recognition method according to the present invention and the speech recognition apparatus that operates according to the procedure defined by the speech recognition method can be configured by hardware. However, in order to realize the simplest, the speech recognition program according to the present invention is installed in a computer. However, the embodiment in which the computer functions as a voice recognition device is the best.
The voice recognition program according to the present invention is written in a computer-readable program language, recorded on a recording medium such as a magnetic disk or CD-ROM that can be read by the computer, and installed in the computer from these recording media or through a communication line. Is done. When the installed program is decoded by a CPU provided in the computer, the computer executes a speech recognition operation in accordance with the procedure shown in FIG.

  The speech recognition method and speech recognition device according to the present invention are utilized in the field of, for example, an automatic guidance device or an automatic reservation device.

The block diagram for demonstrating one Example of the speech recognition apparatus by this invention. The flowchart for demonstrating the procedure of the speech recognition method by this invention. The block diagram for demonstrating the prior art.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Microphone part 19 Telephone 12 Speaker part 20 Branching device 13 Transmission part 21 Received signal maximum amplitude estimation part 14 Reception part 22 Side sound gain estimation part 15 Side sound circuit 23 Side sound suppression process part 16 Transmission signal recording part 110 Speech recognition apparatus
17 Received signal recording unit 111 Speech recognition processing unit 18 Handset

Claims (8)

In a voice recognition method for taking out a transmission signal and a reception signal from a telephone and recognizing these transmission signals and reception signals separately,
Sidetone gain estimation process for estimating the gain that the transmitted signal wraps around the received signal;
A process for estimating the maximum amplitude of the received signal for estimating the maximum amplitude of the received signal;
Using the estimated sidetone gain and received signal maximum amplitude to estimate the sidetone signal for the transmitted signal;
A side-tone suppression process that suppresses the side-tone signal from the received signal based on the estimated side-tone signal;
A speech recognition process for recognizing the received speech and the transmitted speech separately after side-tone suppression processing;
A speech recognition method comprising:   2. The speech recognition method according to claim 1, wherein the side-tone gain estimation is calculated based on a ratio of a reception signal and a transmission signal.   3. The speech recognition method according to claim 1, wherein the maximum amplitude of the received signal is estimated based on the maximum amplitude in the received signal cut out every unit time. In a voice recognition device for taking out a transmission signal and a reception signal from a telephone and recognizing these transmission signals and reception signals separately,
A sidetone gain estimator that estimates the gain that the transmitted signal wraps around the received signal;
A received signal maximum amplitude estimating unit for estimating the maximum amplitude of the received signal;
A side-tone signal estimating unit that estimates a side-tone signal with respect to the transmitted signal using the estimated side-tone gain and the maximum amplitude of the received signal;
A side-tone suppression processing unit that suppresses the side-tone signal from the received signal based on the estimated side-tone signal;
A voice recognition processing unit for recognizing separately the received voice and the transmitted voice after the side-tone suppression processing;
A speech recognition apparatus comprising:   5. The speech recognition apparatus according to claim 4, wherein the side-tone gain estimation unit is configured by calculation means for calculating based on a ratio between the received signal and the transmitted signal.   6. The speech recognition apparatus according to claim 4, wherein the received signal maximum amplitude estimating unit includes recording means for recording an received signal cut out every unit time, and the received signal recorded in the recorded means is included in the received signal. A speech recognition apparatus for estimating a maximum amplitude of a received signal with a maximum amplitude.   A speech recognition program written in a computer-readable program language and causing the computer to function as the speech recognition apparatus according to claim 4.   A recording medium comprising a computer-readable recording medium, wherein the voice recognition program according to claim 7 is recorded on the recording medium.

Images