JPS63303395A - Voice recognition equipment with multi- amplification function - 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] [Field of Industrial Application] The present invention relates to a speech recognition device that recognizes uttered speech, and in particular, a multi-amplification function that reliably extracts phoneme-level features from speech parts with low utterance levels. The present invention relates to a speech recognition device equipped with the following.

[Conventional technology]

Conventionally, in this type of speech recognition device, the amplification degree of the amplifier for amplifying the speech signal from the microphone is determined by the speaker before registration (processing performed by the speech recognition device for a specific speaker) and recognition processing. The voice is automatically set by the voice recognition device according to the volume set by the user himself or the level of the test voice of the speaker (hereinafter, this level setting process is referred to as level setting), and the voice is amplified by the amplification level. Voice recognition processing was performed on the signal.

[Problem that the invention seeks to solve]

In the conventional speech recognition device described above, registration and recognition are performed after setting the level, and the amplification determined by the level setting is not changed during subsequent registration and recognition processing, and the audio signal from the microphone is processed at a constant amplification. Because it amplifies and processes the audio signal, the degree of amplification is small in the low-volume sections during utterance, especially the consonant sections (the quantization accuracy after LΦ conversion is rougher than in the high-volume sections). The disadvantage is that voice recognition has to be performed.

[Means for solving problems]

The speech recognition device equipped with the multi-amplification function of the present invention, when extracting and recognizing phoneme-level features from uttered speech, selects the optimum amplification degree for the time period for extracting the phoneme-level % characteristics. The system reads out a voice signal from the storage section and recognizes the entire voice while extracting features at the phoneme level from the signal.

[Effect]

Since the present invention can extract the features from the audio signal amplified at the optimum level, the recognition rate can be improved.

〔Example〕

FIG. 1 is a block diagram showing an embodiment of a speech recognition device equipped with a multi-amplification function according to the present invention. In the figure, MC is a microphone, Al + A2 to A5 are amplifiers each having different amplification degrees a1 to a5, C,
, C2 to C5 are A/D converters that convert input analog audio signals into digital voice signals and output them, and SD and ED detect the beginning and end of the audio signals, respectively, and output a starting edge detection signal S1 and an ending edge detection signal, respectively. The start edge detection section and the end detection section M1+ which output the signal S2, M2 to M5 start the storage operation upon input of the start edge detection signal S1, and store the digital audio signals output from the converters C7 to C5. The storage unit SE, which completes the storage operation upon input of the termination detection signal S2, selects the audio signal stored at the optimum amplification degree for the frame section from the storage units M, to M5, and stores the audio signal V. The amplification selector RC outputs the volume level of each frame section as a volume level signal, while dividing the sent audio signal V into, for example, several rnsO frames and extracting the phoneme level from each frame. This is a speech recognition unit that extracts features, matches them with standard frame patterns, and outputs a speech recognition result signal T with phoneme labels attached. In addition, Figure 2 shows the volume change of the uttered voice "Asa" (morning) from frame F1 to
It is a diagram divided into F+5. Further, the speech recognition unit RC recognizes, for example, a vowel as a phoneme feature.

g Select a rasp, a nasal, or a plosive, register each standard frame pattern in advance, match it with that standard frame pattern during recognition, and assign a phonological label name to each that indicates the attribute of the standard frame pattern with the highest similarity. attached to the frame.

Next, the operation of the speech recognition device having the multi-amplification function configured as described above will be explained. First, the audio signals input from the microphones MC have different amplification degrees of 1. ~
After being amplified by amplifiers A1 to A5 equipped with a5, ~Φ
input to converters C1-C5. The converters C1 to C5 convert the input analog audio signals into digital audio signals and output the digital audio signals to the storage units M and M5. On the other hand, the start end detection unit SD and the end detection unit ED store the start end detection signal S1 and the end detection signal S2 of the audio signal in the storage units M1 to
Output to M5. Therefore, the storage units M1 to M5 store the digital audio signals output from the Φ converters C1 to C5 from when the start detection signal S1 is input to when the end detection signal S2 is input.

Then, the speech recognition section RC reads out the speech signals from the storage sections M and -M5, extracts features at the phoneme level from the speech signals, and recognizes the speech from the beginning to the end. The speech uttered [Asaj (morning)] is divided into frames of several milligrams each, phoneme-level features are extracted from each frame, and a phoneme label is attached to each frame. That is, the speech recognition unit RC outputs a volume level signal from the volume level of the frame F1 to the amplification degree selection unit SE. Therefore, this amplification selector SE selects the frame F1.
The audio signal stored at the optimum amplification degree for the section is selected from the storage units M1 to M5 and outputted as the audio signal V. Therefore, the speech recognition unit RC extracts features from the input speech signal V, matches them with the standard frame pattern, and attaches a phoneme label. Similarly, phonological labeling can be applied to frames F2 to F+5 as well.

Then, the word dictionary has a phonological label string of the word to be recognized (hereinafter referred to as the word phonological label string), and the phonological label string obtained from the unknown input signal is matched with the word phonological label string, and the similarity is calculated. Word recognition can be performed by using the attribute of the word in the largest word phoneme label string as the recognition result.

The above explanation is based on word recognition as an example, but it is not limited to this; by having a language-level dictionary containing connection information between words in addition to a word dictionary, it is possible to recognize continuously uttered speech as well. Of course, it can be recognized in the same way. Further, although the feature has been described as a phonetic label and the recognition method as matching, it is needless to say that the present invention is not limited to this. Further, in the above description, a case has been described in which one amplifier, one converter, and five storage units are provided, but the present invention is not limited to this, and it goes without saying that any number may be used.

〔Effect of the invention〕

As explained in detail above, the speech recognition device equipped with the multi-amplification function according to the present invention is effective in extracting phoneme-level features from voice parts with low utterance levels, such as the beginning and end of words and consonants. Since the features are extracted from the audio signal that has been amplified at the optimal level for the section, it has the effect of reducing recognition errors and errors.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing an embodiment of a speech recognition device equipped with a multi-amplification function according to the present invention, and Fig. 2 shows changes in the volume of the audio signal from the microphone when the speaker utters "Asa". FIG. MC... e microphone, A1-A5... group amplifier, T... speech recognition result, self-C5... Φ converter, Ml-M5... memory section, SE...・φ amplification selection section, RC...speech recognition section, SD@...starting end detection section, ED...end detection section, Sl...starting end detection signal, S2...end detection Signal, L-...Volume level signal.

Claims (1)

[Claims] A plurality of amplifiers, each having a different amplification degree, amplify the output of the microphone, a memory section that temporarily stores the audio output signals of these amplifiers from the beginning to the end, and an optimal 1. A speech recognition device having a multi-amplification function, comprising means for reading an amplification-level speech output signal from a storage unit and performing recognition processing.