JPS6091397A - 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 The present invention relates to a speech recognition device that can efficiently recognize speech.

In the Saramai Giken voice recognition device, when voice input overlaps with one recognition device, conventionally, it is temporarily recorded on a recording medium such as a recorder, and then played back when there is free time and recognized sequentially. However, this method has the disadvantage that it takes time to reproduce and the amount that can be recognized is limited.

Purpose The present invention has been made in order to solve the above-mentioned drawbacks of the prior art.In particular, in a voice recognition device, when a large number of inputs overlap at the same time, the voice recorded by means such as recording is recorded. This was done with the aim of providing a speech recognition device that can reproduce data faster and recognize it more efficiently.

Configuration The configuration of the present invention will be described below based on examples.

The present invention was made in view of the above-mentioned circumstances, and
In particular, by increasing the playback speed of the recorded voice and playing it back faster than when it was recorded, the input voice can be recognized in a time-efficient manner. The following describes the case where the image is played back at double speed and recognized.

The figure is an electrical block diagram for explaining one embodiment of the present invention, and in the figure, l represents a microgon.

2 is means for playing back what was recorded on a recorder at double speed, 3 is a switch for selecting these two inputs, 4 is a pre-emphasis circuit for correcting the acoustic attenuation characteristics of human voice, and 5 is a characteristic parameter. A B-P-F section for extracting a certain power spectrum using, for example, a B-P-F (band pass filter) group; 6 and 7 are a detection section and a smoothing section for extracting power from the output of the filter 5; This smoothing section is composed of L-, P, and F (low-pass filters). 8
9 is an AD converter that quantizes the power spectrum; 9 is a switch that switches and inputs the feature pattern consisting of the quantized power spectrum to either the dictionary section 10 or the collation section 11 in the next stage; The matching section 11 is provided to store patterns, and the matching section 11 is provided to match the input speech with characteristic patterns in the dictionary 10. 12 is a determination unit that determines the content of the input audio based on the comparison result;
Reference numeral 4 denotes a control section that controls the setting parameters of the pre-emphasis circuit 4 to the AD converter 8 to double speed or normal speed using the switching signal 13 from the switch 3.

Table 1 shows the parameters for normal speed and double speed.
In the case of double speed, the reproduction time is half the recording time, and the frequency of the reproduced signal is shifted twice. Therefore, the setting parameters are set so that the signal shifted twice is processed.

Table 1 As is clear from Table 1, pre-emphasis circuit 4;
-P-F group 5; L-P-, F section 7; and AD
Each setting value of the converter 8 is twice the normal setting value. However, the frame time will be half of the normal time.

If the settings are made as described above, the data amount will not change as a whole because the number of samples will double, but the playback time will be halved. Therefore, there is a one-to-one correspondence between the characteristic pattern data registered in the dictionary 10 as normal and the data processed at double speed, and N matching can be performed as is.

Note that although the explanation has been made using double speed as an example, the present invention is not limited to the above embodiment; for example,
When the playback speed is double, the playback time and frame time are 1
It goes without saying that by multiplying /n by n times the other parameters in Table 1, n times more recognition can be achieved.

In addition, although the example above uses the B-H-F group for audio capture, the parameter settings are the same when input audio is immediately quantized and operations after pre-emphasis are performed using a digital filter, etc. , and characteristic parameters other than those explained above (e.g. autocorrelation coefficient, P
It goes without saying that recognition using (arcor coefficients, cloud intersection numbers, etc.) can be processed in exactly the same way.

As is clear from the above explanation, according to the present invention, even if the recorded audio is played back at a faster playback speed than the recording speed, the registered dictionary and audio capture setting parameters can be changed. It is possible to provide a speech recognition device that performs accurate speech recognition by performing the same verification operation as in the normal case, and it is possible to perform efficient recognition.

[Brief explanation of the drawing]

The figure is an electrical block diagram for explaining one embodiment of the present invention. DESCRIPTION OF SYMBOLS 1... Microphone, 2... Reproduction device, 4... Pre-emphasis circuit, 5... B-P-F group, 6... Detection section, 7... Smoothing section (L-P-F), 8... A/D converter, 10... Dictionary section, 11... Verification section, 12...
Judgment unit, 14...control unit.

Claims (3)

[Claims] (1) an extraction unit that extracts the feature parameters of the audio signal; a dictionary unit that stores the feature parameters of the audio to be registered;
A speech recognition device that has a matching section that matches the feature parameters of the input speech with the feature parameters stored in the dictionary section, and a determination section that discriminates the input speech from the matching result, uses direct speech input and speech that has been recorded once. A voice recognition device characterized by switching between reproducing a signal and applying a voice input to recognize each one. (2) The speech recognition device according to claim (1), further comprising a control means for obtaining a control signal from a selection signal for switching between direct speech input and recorded speech input. (3) The speech recognition device according to claim (2), characterized in that the extraction parameters of the feature extraction section are changed depending on the playback speed (n times) of the recorder by a control signal from the control means. .