JPH041915B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Field of Industrial Application The present invention relates to a speech recognition device that recognizes speech.

(b) Prior Art Figure 1 shows the configuration of a conventional speech recognition device. In the figure, 1 is a microphone that converts audio into an electrical audio signal; 2 is an audio pattern extraction circuit that extracts a audio pattern representing the characteristics of audio from the audio signal obtained by the microphone 1; and 3 Reference numeral 4 designates an input speech pattern memory that stores the speech obtained by the speech pattern extraction circuit 2, and 4 indicates a registered speech pattern memory that stores in advance numbered speech patterns of a plurality of speeches to be recognized. 5 is a recognition processing unit that performs pattern recognition, which compares the voice pattern in the input voice pattern memory 3 with each voice pattern in the registered voice pattern memory 4, and determines the number of the most similar voice pattern in the registered pattern memory 4. is output. That is, the recognition processing of the voice input to the microphone 1 is performed by the basic configuration consisting of the microphone 1 to the recognition processing section 5 described above. Furthermore, 6 is a comparison circuit, 7 is a counter, 8 is a display, and 9 is a correction section. The comparison circuit 6 to the correction section 9 perform processing for correcting each voice pattern in the registered voice pattern memory 4 to a more accurate pattern.

Next, the operation will be described by taking as an example a case where the conventional speech recognition device as described above is used to recognize a place name speech.

First, in the registration mode, the speaker sequentially inputs place name sounds into the microphone 1. At this time, the counter circuit 7 sequentially increments the numerical display on the display 8 from "1" until the numerical display reaches "1".
At this time, the speaker inputs the voice "Tokyo" into the microphone 1. At this time, the voice pattern "TOKYO" extracted by the pattern extraction circuit 2 is temporarily stored in the input pattern memory 3, and then this voice pattern is associated with the number "1" in the registered voice pattern memory 4. and store it. Next, when the number display on the display 8 increments to "2", the speaker inputs the voice "Nagoya" into the microphone 1, and the voice pattern "Nagoya" is similar to the case of "Tokyo" described above. is stored in the registered voice pattern memory 4 in association with the number "2". Similarly, as the speaker utters, the voice patterns ``Kiyouto'', ``Osaka'', ``Okayama'', ``Hiroshima'', and ``Hakata'' are sequentially stored in the registered voice pattern memory 4 with numbers ``3'' and ``Hakata''. 4”, “5”, “6”, “7”
Stored in association with

However, in the registration mode described above, it is not always possible for the speaker to pronounce with correct pronunciation and accent, and there is a risk that an incorrect speech pattern may be stored in the registered speech pattern memory 4. A modification mode is provided for modifying the voice pattern. In this correction mode, the counter circuit 7 sequentially increments the numerical display on the display 8 from "1" to "7" in the same manner as in the registration mode, and when the numerical display is "1", the speaker inputs the voice "Tokyo" into microphone 1.
At this time, the voice pattern "TOKYO" extracted by the pattern extraction circuit 2 is temporarily stored in the input voice pattern memory 3. Then, the recognition processing unit 5 recognizes the voice pattern "TOKYO" in the input voice pattern memory 3 based on each of the seven voice patterns in the registered voice pattern memory 4. As a result, "Tokyo" is associated with the number "1" in the registered voice pattern memory 4.
If the voice pattern is correctly recognized as a voice pattern, the number “1” is output from the recognition processing unit 5.
The comparator circuit 6 detects that the count value "1" of the counter circuit 7 and the count value "1" of the counter circuit 7 match, and outputs a match signal S. This coincidence signal S increments the count value of the counter circuit 7, and changes the number displayed on the display 8 to "2". Thereafter, in the same way, according to the numerical display on the display 8, place names corresponding to the numbers are input into the microphone 1 in sequence. For example, if the numerical display on the display 8 is "4", "Osaka" When a voice is input into the microphone 1, the recognition processing unit 5 determines that the input voice pattern at this time is the registered voice pattern that is most similar to the voice pattern ``Okayama'' corresponding to number "5" in the registered voice pattern memory 4. If it is determined that there is one and the number "5" is output, the comparison circuit 6 which compares this number "5" with the count value "4" of the counter circuit 7 outputs a mismatch signal. This mismatch signal causes the correction section 9 to operate without causing the counter circuit 7 to step forward. At this time, the correction unit 9 uses the current "Osaka" stored in the input voice pattern memory 3.
A new averaged speech pattern of "Osaka" is obtained by averaging the speech pattern of "Osaka" and the speech pattern of "Osaka" associated with the number "4" of the registered speech pattern memory 4, and the speech pattern is stored in the registered speech pattern memory. The voice pattern "Osaka" in No. 4 is changed to this new averaged voice pattern.
At this time, the number displayed on the display 8 remains "4", and the speaker again inputs the voice "Osaka" into the microphone 1, and the recognition processing unit 5 accurately records the voice "Osaka". The operation for correcting the voice pattern "Osaka" in the registered voice pattern memory 4 as described above is repeated until the number "4" is output. When this modification is completed, the process moves on to the above-mentioned modification process for the voice "Okayama" in the next registered voice pattern memory 4, and this process is performed for the last voice "Hakata".
This will be carried out in sequence.

In the recognition mode, the voice pattern input to the microphone 1 is extracted 2 and stored in the input voice pattern memory 3, and is processed by the recognition processing unit 5 based on each voice pattern in the registered voice pattern memory 4.
The pattern is recognized. That is, if the recognition processing unit 5 outputs the number "4", the voice uttered by the speaker at this time was "Osaka".

However, in such conventional speech recognition devices, when in the registration mode, the pronunciation and accent of the voice "okayama" may be inaccurate, or noise may be mixed in when the voice is uttered, resulting in errors in registration. The voice pattern "Okayama" corresponding to the number "5" in the voice pattern memory 4 may become similar to the voice pattern "Osaka" corresponding to the number "4".
In this case, in the modification mode, the voice pattern "Osaka" in the registered voice pattern memory 4 is first modified according to the order of the associated numbers, but "Osaka" corresponding to this number "4" is corrected first. Even if the sound pattern ``Osaka'' uttered by the speaker is quite accurate, the sound pattern ``Osaka'' uttered by the speaker may be rather similar to the inaccurate sound pattern ``Okayama'' that corresponds to the number "5". be. At this time, the recognition processing unit 5 outputs the number "5" and corrects the voice pattern "Osaka" of the registered voice pattern 4.
It is not possible to perform a correction process that can sufficiently distinguish the sound pattern "Osaka" from the similar sound pattern "Okayama". Therefore, not only is it impossible to complete the correction process for the voice pattern "Osaka" in the registered voice pattern memory 4, but also
There was a drawback that subsequent correction processing of the voice pattern was no longer possible.

(c) Purpose of the invention The purpose of the present invention is to eliminate the above-mentioned drawbacks,
An object of the present invention is to provide a speech recognition device that can reliably modify speech patterns in a registered speech pattern memory.

(d) Structure of the Invention The speech recognition device of the present invention includes a modification means for modifying each speech pattern in the registered speech pattern memory;
and instruction means for instructing the speaker to utter the registered voice to be corrected by the correction means, and in the correction mode, when the speaker inputs the registered voice instructed according to the instruction means, other registered voices are input. If the correcting means misrecognizes the corrected registered speech, the correcting means changes the target of the registered speech to be corrected to the misrecognized registered speech, and
The above-mentioned instruction means is for instructing the speaker of the newly registered voice changed by the modification means.

(E) Embodiment FIG. 2 shows an embodiment of the speech recognition device of the present invention. In this figure, parts having the same functions as those of the conventional device shown in FIG. 1 are given the same numbers as in FIG.
Note that 6', 8', and 9' indicate a comparison circuit, a display, and a correction unit as in the conventional device, but as will be described later, the objects for comparison, display, and correction are different from those in the conventional device. different. 10 is a selection circuit that selectively outputs the count value of the counter circuit 7 and the output number of the recognition processing section 5, and transmits this output value to the comparison circuit 6', the display 8', and the correction section 9'; When receiving the match signal S from the circuit 6', the count value of the counter circuit 7 is output. Conversely, when a mismatch signal is received from the comparison circuit 6', the output number from the recognition processing section 5 is output, and the next time either the match signal S or the mismatch signal is received from the comparison circuit 6'. This output value is then returned to the count value of the counter circuit 7. Reference numeral 11 denotes an opening/closing circuit for opening and closing the connection between the counter circuit 7 and the coincidence signal S line that causes the counter circuit 7 to increment. The coincidence signal S is connected to the counter circuit 7 only when a count value of 7 is output. Therefore, when the output number from the recognition processing section 5 and the output value from the selection circuit 10 match, the comparison circuit 6' outputs a match signal S, and conversely, when they do not match, it outputs a mismatch signal. , the opening/closing circuit 11 is brought into a connected state by this coincidence signal S, thereby causing the counter circuit 7 to step forward, and by the disagreement signal being caused to execute the correction operation of the correction section 9', furthermore, these two signals S , causes the selection circuit 10 to perform a selection operation. Further, the display section 8' displays the count value of the counter circuit 7 from the selection circuit 10 or the output number from the recognition processing section 5. Further, when the modification section 9' receives the discrepancy signal from the comparison circuit 6', it modifies the speech pattern in the registered speech pattern memory 4 corresponding to the output number of the recognition processing section 5 obtained from the selection circuit 10. Become.

The operation when the speech recognition device of the present invention having such a configuration is used to recognize a place name speech will be described. Note that the registration mode and the recognition mode operate in the same manner as the conventional device, so their explanation will be omitted.

Therefore, "Tokyo", ~, "Hakata" are stored in the registered voice pattern memory 4 as in the conventional example.
The modification mode will be described below, assuming that the voices are stored in association with the numbers "1" to "7".

First, in the initial state, the count value "1" of the counter circuit 7 is forcibly input to the display 8' via the selection circuit 10, and the display 8' displays the number "1". In accordance with this, the speaker inputs the voice "Tokyo" associated with the number "1" into the microphone 1, and this voice pattern is extracted as a pattern 2 and temporarily stored in the input voice pattern memory 3. , the pattern is recognized by the recognition processing section 5. At this time, if the recognition processing unit 5 correctly determines that the voice pattern is "TOKYO" which is associated with the number "1" in the registered voice pattern memory 4, it will output the number "1" and the comparison circuit 6' is the opening/closing circuit 11 in which the coincidence signal S is closed.
The signal is inputted to the counter circuit 7 and transmitted to the selection circuit 10 via. Therefore, the correction section 9' does not perform any correction operation, and the selection circuit 10 selects the incremented count value "2" of the counter circuit 7 and displays it on the display 8'.
to be displayed. In this way, the numeric display on the display 8' is sequentially changed in the order of the corresponding numbers until a non-coincidence signal is obtained from the comparison circuit 6', and the speaker sequentially inputs the place name speech into the microphone 1 accordingly.

On the other hand, for example, the number displayed on the display 8' is "4".
When the corresponding voice "Osaka" is input into the microphone 1, the recognition processing unit 5 recognizes that the voice pattern at this time corresponds to number "5" in the registered voice pattern memory 4 for some reason. If it is determined that the voice pattern is most similar to the incorrect voice pattern “Okayama” and the number “5” is output,
Comparison circuit 6' outputs a mismatch signal. In this case, the switching circuit 11 is opened in response to the mismatch signal, and the counter circuit 7 maintains the count value "4". On the other hand, the selection circuit 10 selects the number "5" from the recognition processing section 5 at this time and displays it on the display 8', and furthermore, the modification section 9' performs a modification operation. That is, the modification unit 9' modifies the registered voice pattern memory corresponding to the number immediately before the output from the selection circuit 10 is changed to the number "5", that is, the count value "4" held by the counter circuit 7. The first voice pattern is “Osaka”.
Similar to the modification section 9 of the conventional device shown in the figure, after changing to the averaged speech pattern, the modification section 9' changes the "OKAYAMA" pattern in the registered speech pattern memory associated with the number "5" obtained from the selection circuit 10. ” sound pattern is subject to new modification.

Thus, when the speaker inputs the voice "Okayama" into the microphone 1 according to the number "5" on the display 8', the registered voice pattern memory 4 is changed by the operation of the correction section 9'. The voice pattern "Okayama" is corrected to the averaged pattern,
At this time, the output number as the recognition result of the recognition processing unit 5 is not necessarily "5". That is, since the voice pattern "Okayama" in the registered voice pattern memory 4 is inaccurate, the output number is often not "5". Therefore, from the comparator circuit 6',
Either the match signal S or the mismatch signal is obtained, and depending on this signal S+, the switching circuit 11 is closed, and the selection circuit 10 again resets the count value "4" held in the counter circuit 7 until now. Output. As a result, the correction target of the correction unit 9' returns to the voice "Osaka", and the number displayed on the display 8' is changed from "5" to "4".
The speaker is instructed to input the voice "Osaka" into the microphone 1 again.

In this way, the inaccurate voice pattern of the voice "Okayama" in the registered voice pattern memory 4, which is often misrecognized as the voice "Osaka", is first corrected, and then the voice pattern of "Osaka" is changed again. Correction processing is started for the However,
The correction process at this time is the registered voice pattern memory 4.
The sound pattern "Okayama" has already been modified to the extent that it can be distinguished from the sound pattern "Osaka", and this sound pattern "Osaka" has also been modified once, so this modification process If this is repeated at most two or three times, the influence of inaccurate speech patterns other than the "Osaka" speech pattern will disappear, and the correction of the "Osaka" speech pattern can be completed. Therefore, the comparison circuit 6' outputs the match signal S.
is output, and it is possible to quickly move on to the modification process for the next number "5", which is the sound "Okayama", and in the same way, the modification process is sequentially performed up to the last number "7", which is the sound "Hakata". It will be done.

In the above explanation, as a modification means for modifying the registered speech pattern in the registered speech pattern memory 4, the modification section 9' for modifying the registered speech pattern using an averaging pattern has been shown. It is also possible to simply change the input voice pattern to the input voice pattern.

In addition, as an instruction means for instructing the speaker which voice to utter, a display device 8' that displays a number associated with the voice is used, but the voice may be displayed as is. It is also possible to provide a voice synthesis circuit and output voice using this circuit.

(f) Effects of the Invention As is clear from the above description, the speech recognition device of the present invention includes a modification means for modifying each speech pattern in the registered speech pattern memory, and an utterance of the registered speech to be modified by the modification means. and instructing means for instructing the speaker, and in the correction mode, when the registered voice instructed by the speaker according to the instruction means is input and it is mistakenly recognized as another registered voice, the correction means changes the target of the registered voice to be corrected to the erroneously recognized registered voice, and the above-mentioned instruction means instructs the speaker to the new registered voice changed by the correction means. Even if one voice pattern is relatively accurate and another voice pattern is mistakenly similar to the previous voice pattern for some reason, the first voice pattern can be corrected after correcting this inaccurate other voice pattern. Voice patterns can be modified. Therefore, it is possible to correct the previous speech pattern accurately to the extent that it can be sufficiently distinguished from other speech patterns, and such correction processing can be completed reliably and quickly, making it easier to operate this type of speech recognition device. Not only can this be expected to greatly improve performance, but it also contributes to reducing erroneous recognition in recognition mode.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a conventional speech recognition device, and FIG. 2 is a block diagram of an embodiment of the speech recognition device of the present invention, where 1 is a microphone, 2 is a speech pattern extraction circuit, and 3 is an input speech pattern. Memory, 4 is a registered voice pattern memory, 5 is a recognition processing unit, 6 and 6' are comparison circuits, 7 is a counter circuit, 8 and 8' are display devices,
Reference numerals 9 and 9' indicate correction sections, 10 a selection circuit, and 11 an opening/closing circuit, respectively.

Claims (1)

[Scope of Claims] 1. A recognition processing unit converts a voice pattern obtained by a pattern extraction circuit from a voice signal obtained by inputting voice into a microphone into a pattern based on a plurality of voice patterns in a registered voice pattern memory. In a speech recognition device that performs recognition, an instruction means for sequentially instructing registered speech to be reuttered when a speaker faces a microphone in order to modify each speech pattern in a registered speech pattern memory;
correction means for correcting the voice pattern of the registered voice instructed by the instruction means based on the voice pattern extracted by the pattern extraction circuit from the voice reuttered by the speaker when facing the microphone; In the pattern correction mode, if the recognition processing results for the instruction registration voice instructed by the instruction means and the voice input by the speaker into the microphone according to the instruction are different, the input voice pattern at this time is The modification means modifies the voice pattern of the registered instruction voice based on the above, and the instruction means instructs the registered voice using the registered voice pattern indicated by the recognition processing result as a new correction target. Device.