JPS60205600A - Voice recognition equipment - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a speech recognition device used, for example, in a speech recognition response system.

[Technical background of the invention and its problems] A voice recognition response system is used, for example, in a system that exchanges information between a user and a host computer using a telephone line and provides information services to the user. .

Such a voice recognition response system is a speaker-independent wgi
It is configured to recognize code numbers, mainly consisting of numbers, uttered by the user using a voice recognition device.

In this case, the user has to utter a code number consisting of a series of numbers, dividing it into words and uttering them continuously. The speech recognition device regards each number uttered by the user as a word and recognizes each number word by word.

By the way, when a voice recognition device recognizes a voice from a user, it is necessary to detect an acoustic section to be recognized from a voice signal input from a telephone line. The voice detection method for detecting this voice section involves statistically determining a certain detection level L for the voice signal U, as shown in FIG. There is a method for detecting E. However, the audio level of audio signals transmitted via a telephone line may vary greatly depending on the line condition. For this reason, in the voice detection method described above, when the level of the noise N is high, a situation occurs where the noise N exceeds the detection level L as shown in Figure 2 (a), making it impossible to detect the voice section. become. Furthermore, as shown in FIG. 2B, the level of the audio signal U may fall below the detection level L, making it impossible to detect the audio section.

In contrast to the above-mentioned methods, a voice detection method has been developed in which the detection level L is changed depending on the level of the background noise N. However, in the case of this method, as shown in FIG. 3, when the background noise level is low and the audio level is relatively high, the set detection level L will be relatively low.

For this reason, there is a drawback that the breathing of the speaker (user) or small noises in the surroundings are sensitively detected. Furthermore, when detecting the end E of the voice in the voice detection process, as shown in FIG. It is necessary to determine whether the process has ended. It is difficult to set this time (hereinafter referred to as audio duration) T to an appropriate value in advance. For example, the audio duration T
If is set to a relatively short value, a situation may occur in which the detection of the speech section of a word with a relatively long utterance ends before the original end is detected. Also, if the audio duration T is made relatively long,
This method has the drawback that it takes a lot of time to detect a voice section, and the efficiency of response processing deteriorates.

[Object of the Invention] The present invention has been made in view of the above-mentioned points, and its object is to provide an appropriate detection level and a suitable detection level even when the audio level fluctuates with respect to an audio signal input via a telephone line. An object of the present invention is to provide a speech recognition device capable of performing highly accurate speech recognition processing by setting a speech duration and reliably detecting a speech section.

[Summary of the Invention] The present invention provides a voice detection means for detecting a voice section consisting of a start and end at a predetermined detection level in a voice signal transmitted via a telephone line, and temporarily storing the waveform locus of the voice signal. is provided. The voice LL1g device performs voice recognition processing based on the voice section detected by the voice detection means. Furthermore, the voice detection control means detects the maximum level of the voice signal and the background noise level based on the waveform locus of the voice signal stored by the voice detection means, sets the appropriate detection level, and sets the voice duration. The controller is configured to perform control to detect and determine the end of the voice section. With the speech recognition device having such a configuration, it is possible to reliably detect a speech section in a speech signal in which fluctuations in speech level are likely to occur, and it is possible to perform speech recognition processing at a high $1 degree.

[Embodiment of the Invention] An embodiment of the present invention will be described below with reference to the drawings. Fourth
The figure is a block diagram showing the configuration of a voice recognition response system according to an embodiment. In FIG. 4, a telephone set 10 converts the voice uttered by a user into an electrical signal and connects it to a telephone line 1.
1 to the voice detection circuit 12. The voice detection circuit 12 detects the voice section, that is, the start end S and the end point E of the voice based on a predetermined detection level L of the transmitted voice signal.

The audio detection circuit 12 also temporarily stores the waveform locus of the initial audio signal (the audio signal including the background noise N before and after) in the memory 13. The voice recognition circuit 14 is the voice detection circuit 12
Recognition processing is performed on the voice in the voice section detected in . □ The control circuit 15 reads out the waveform trajectory information of the audio signal stored in the memory 13 by the audio detection circuit 12, and detects the maximum level of the audio signal and the background noise level based on this information. Based on this detection, the control circuit 15 sets the detection level L necessary for detecting the voice section to an appropriate value in the voice detection circuit 12.

Furthermore, the control circuit 15 detects the voice duration based on the waveform locus information, and controls the voice detection circuit 12 to determine the end point E when the voice detection circuit 12 detects the next voice section.
Controls the operation of 2. The voice response device 16 includes a control circuit 1
5 performs response processing according to the recognition result of the speech recognition circuit 14.

The operation of one embodiment of the voice recognition response system configured as described above will be described. First, when the user operates the telephone 10 to establish a line with the system, a predetermined first message is sent from the voice response device 16 to the user. This first message could be, for example, “×
First, I would like to tell you about the service of ×. Please say yes if you like. ”. In response, the user answered "Yes"
When uttered, the uttered voice is transmitted to the voice detection circuit 12 via the telephone line 11.

As shown in FIG. 5(a), the audio detection circuit 12 detects the starting end S and the ending end E of the input audio signal U at a predetermined detection level L, and instructs the control circuit 15 to detect them.

At this time, the audio detection circuit 12 stores the waveform locus of the first audio signal (including the background noise N) U in the memory 13. The voice detected by the voice detection circuit 12 is subjected to voice recognition processing by the voice recognition circuit 14, and the recognition result is sent to the control circuit 1.
5 is output.

In this case, when the voice from the user is determined to be noise by the voice detection circuit 12, the voice response device 16 sends a message to the user under the control of the control circuit 15, such as "Cannot confirm. Please try again." Sent.

The control circuit 15 controls the memory 13 through the voice detection circuit 12.
The waveform locus information of the audio signal U is read from the audio signal U, and the maximum audio level and background noise level are detected based on the waveform locus information. According to this detection result, the control circuit 15 determines an appropriate detection level L for detecting the voice section and sets it in the voice detection circuit 12. Further, the control circuit 15 determines the voice duration T1 as shown in FIG. 5(a) from the waveform trajectory information.

, Next, when the user utters a series of code numbers, the uttered voice signal is input to the voice detection circuit 12 in the same manner as described above. Here, the voice detection circuit 12 detects the voice section based on the appropriate detection level L set by the control circuit 15. At this time, the voice detection circuit 12 can reliably detect the end E of the voice section based on the voice duration T1 determined by the control circuit 15. That is, even if the audio level fluctuates and falls below the detection level L, if the time for the level drop is equal to or less than the audio duration time T1, it is determined that the audio continues. This audio duration T1 is fixed until the series of information services is completed. This is because it is considered that the duration of voice when uttered by the same user is almost the same.

Therefore, when the control circuit 15 detects the voice duration T2 as shown in FIG. 5(b) by uttering the first message depending on the user, this voice duration T2 is fixed. become.

In this way, from the user's initial utterance in response to the first message, the appropriate detection level L and voice duration used for detecting the voice section are set. therefore,
Even if the level of the voice signal varies somewhat depending on the state of the telephone line, the voice detection circuit 12 reliably detects the voice section. Furthermore, since the voice duration is set by the user's initial utterance, the voice duration can always be fixed at an appropriate value, and the end of the voice section can be reliably determined. This results in
use! ! The voice recognition circuit 14 accurately converts the series of code numbers sent via the telephone line into i! can be understood.

[Effects of the Invention] As described in detail above, according to the present invention, even when the level of a voice signal transmitted via a telephone line fluctuates, the voice section of the voice can be reliably detected. Therefore, for example, a series of code numbers transmitted by a speaker via a telephone line can be voice recognized with high accuracy. As a result, if the present invention is applied to a voice recognition response system, a reliable information response service can be realized.

[Brief explanation of drawings]

1 to 3 are waveform diagrams of audio signals for explaining the operation of conventional speech recognition methods, respectively, and FIG. 4 is a block diagram showing the configuration of a speech recognition response system according to an embodiment of the present invention. Figure 5(a). (b) is a waveform diagram of an audio signal for explaining the operation of the system of FIG. 4, respectively. 11...Telephone line, 12...Voice detection circuit, 13.
...Memory, 14...Speech recognition circuit, 15...1 circuit. Applicant's agent Patent attorney Takeshi Suzue Figure 2 Figure 3

Claims (1)

[Scope of Claims] A voice detection means for detecting a voice section at a predetermined detection level and voice duration from a voice signal transmitted via a telephone line and temporarily storing the waveform locus of the voice signal, and this voice detection means. a voice recognition means that performs voice recognition processing based on the voice signal of the voice section detected by the voice recognition means; and a maximum level of the voice signal based on the voice signal stored by the voice detection means. and voice detection control means for detecting a background noise level and voice duration, setting the appropriate detection level and voice duration, and controlling the voice detection means to detect an appropriate voice section. A speech recognition device characterized by: