JPH0459637B2 - - Google Patents

Description

[Detailed description of the invention] (b) Industrial application fields The present invention relates to a speech recognition device that recognizes speech.

(b) Prior Art The configuration of a conventional speech recognition device is shown in FIG. In the figure, 1 is a microphone that converts audio into an electrical audio signal, 2 is a pattern creation circuit that frequency-analyzes the audio signal and obtains an audio pattern consisting of a time series of spectral values, and 3 is the audio signal. This is a buffer memory that temporarily stores voice patterns obtained from the pattern generation circuit 2. Reference numeral 4 denotes a registered voice pattern memory, in which the voice patterns stored in the buffer memory 3 during the voice registration mode are sequentially introduced via the mode switch Sm connected to the t side and stored as a plurality of registered voice patterns. Reference numeral 5 denotes a pattern recognition circuit, in which a voice pattern obtained through a mode switch Sm connected to the n side from the buffer memory 3 during the voice recognition mode is determined based on each registered voice pattern in the registered voice pattern memory 4. The most similar registered speech pattern is detected.

Therefore, when in voice registration mode, the mode switch
With the Sm connected to the T side, a specific speaker inputs, for example, 68 monosyllables into the microphone 1 in sequence, and these speech patterns 2 are created and stored in the buffer memory 3. After that, it is stored in the registered voice pattern memory 4. Then, in the voice recognition mode, with the mode switch Sm connected to the
The pattern recognition circuit 5 performs pattern recognition based on each registered voice pattern in the registered voice pattern memory 4, and determines that the input voice at this time is "Sa" by detecting the registered voice pattern "Sa". be done.

Such conventional speech recognition devices targeted at unspecified speakers have problems such as ambient noise being mixed in when inputting the registered speech to the microphone 1 in the speech registration mode that precedes the speech recognition mode. Because the speakers' pronunciations are ambiguous and inconsistent, it is not possible to obtain an accurate registered speech pattern for a particular speaker. This method had the disadvantage that speech recognition became impossible and the recognition rate decreased.

(c) Object of the Invention The present invention has been made in view of the above points, and an object thereof is to provide a speech recognition device that prevents a decrease in recognition rate by obtaining accurate registered speech patterns.

(d) Structure of the Invention The speech recognition device of the present invention includes a registered speech pattern memory that stores registered speech patterns of specific speakers and a standard speech pattern memory that stores standard speech patterns of unspecified speakers. mode, the voice pattern of a specific speaker is compared with the standard voice pattern memory standard voice pattern of the same voice, and only when the comparison error is within the allowable range,
This particular private person's voice pattern is stored in a registered voice pattern memory.

(E) Embodiment FIG. 2 shows the configuration of an embodiment of the speech recognition device of the present invention. In the figure, numerals 1 to 5 indicate a microphone to pattern recognition circuit having the same configuration as the conventional device shown in FIG. Reference numeral 6 denotes a standard speech pattern memory, which stores standard speech patterns of, for example, 68 monosyllables representing the characteristics of the speech of unspecified speakers, that is, a large number of speakers. 7 is a comparison circuit which converts buffer memory 3 to mode switch Sm in voice registration mode.
The voice pattern of the input voice obtained through the input voice is compared with the standard voice of the same voice stored in the standard voice pattern memory 6, and when the comparison error is within the permissible range, an allowable signal is output. Reference numeral 8 denotes a gate circuit, which opens the gate circuit 8 only when receiving a permission signal from the comparison circuit 7 and transfers the buffer memory 3 to the buffer memory 3.
The voice pattern of the input voice obtained from the mode switch Sm is introduced and stored in the registered voice pattern memory 4 as a registered voice pattern.

Next, the operation of such a speech recognition device will be described. First, when a specific speaker uses such a device, he or she connects the mode switch Sm to the t side and sets the voice registration mode, and sequentially inputs 68 syllables into the microphone 1. For example, a voice pattern called "sa" is created and stored in the buffer memory 3. Therefore, the comparison circuit 7 compares the voice pattern of the buffer memory 3, which is the "sa", with the upper part of the standard voice pattern memory 6.
At this time, if the specific speaker accurately utters the sound "sa" and an accurate speech pattern is obtained without the adverse effects of ambient noise, the comparison circuit The comparison error in 7 is only the error due to individual differences among specific speakers,
At least the characteristics of the voices are the same, so the error will be within a certain tolerance range, so
The comparator circuit 7 opens the gate circuit 8 with the permission signal.
On the other hand, if an accurate voice pattern cannot be obtained due to ambiguity in the utterance of a specific speaker or the negative influence of ambient noise, the comparison circuit 7 Since the comparison error becomes extremely large and exceeds the permissible range, a permissible signal cannot be obtained, so that the gate circuit 4 remains closed and some display means (not shown) is used. Instructs the specific speaker to input the sound "sa" again.Therefore,
This specific speaker will try inputting the voice "sa" into microphone 1 under better conditions, taking into consideration the quality of pronunciation and the influence of ambient noise, and will be able to obtain an accurate speech pattern. This voice input is performed until the voice is obtained. As a result, a particular speaker's
Accurate speech patterns for all 68 syllables will be stored in the registered speech pattern memory 4 as registered speech patterns.

In the next voice recognition mode, mode switch Sm is connected to the n side in the same way as in the conventional device, and a specific speaker inputs an arbitrary voice into microphone 1 to create a pattern 2, and this voice pattern is is stored in the buffer memory 3, and pattern recognition is performed based on each registered speech pattern of accurate 68 syllables in the registered speech pattern memory 4 stored in the speech registration mode as described above. Therefore, if the voice pattern in the buffer memory 3 is recognized as a registered voice pattern that accurately shows the characteristics of the voice "sa", for example, the input voice at this time has a very high probability, that is, it is highly recognized. It is determined that it is “Sa” based on the percentage.

(f) Effects of the Invention As is clear from the above description, the speech recognition device of the present invention, when registering speech, can distinguish between the speech pattern of a specific speaker to be registered and the same speech pattern of an unspecified speaker. Since it is used as a registered speech pattern only when the comparison error is within an allowable range, it is possible to obtain an accurate registered speech pattern that is free from ambiguity in the pronunciation of a specific speaker and the negative effects of ambient noise. . This increases the precision of pattern recognition processing performed based on these registered speech patterns during speech recognition, and can be expected to greatly improve the recognition rate.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the structure of a conventional speech recognition device, and FIG. 2 is a block diagram showing the structure of an embodiment of the speech recognition device of the present invention, where 2 is a pattern generation circuit, and 4 is a registered speech pattern. Memory, 5 is a pattern recognition circuit, 6 is a standard voice pattern memory,
Reference numeral 7 indicates a comparison circuit.

Claims (1)

[Claims] 1 Create a voice pattern that represents the characteristics of a specific speaker's voice based on the voice of the specific speaker, and recognize this voice pattern based on registered voice patterns of the specific speaker that have been stored in advance. In addition to the registered voice pattern memory for storing registered voice patterns of specific speakers, the voice recognition device has a standard voice pattern memory that stores standard voice patterns representing voice characteristics of unspecified speakers. In preparation, when storing the registered voice pattern of a specific speaker in the registered voice pattern memory in advance, the voice pattern of the specific speaker and the same voice of an unspecified speaker stored in the standard voice pattern memory are stored in advance. A speech recognition device that compares a speech pattern of a specific speaker with a standard speech pattern and stores the speech pattern of a specific speaker in a registered speech pattern memory only when the comparison error is within an allowable range.