JPH0769713B2 - Speech recognition response device - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition and response device that causes a machine to recognize a human voice and causes the machine to respond.

2. Description of the Related Art In recent years, a voice recognition response device using a telephone or the like has been actively developed. In voice recognition for such unspecified speakers, noise-containing voice signals and word heads,
Although it is also necessary to recognize speech signals whose utterances at the end of speech are unclear, the applicant of the present invention has filed a patent application for a speech recognition method capable of obtaining a high recognition rate for these unclear speech signals. Proposed as No. 60-251360. Hereinafter, a voice recognition response device which uses this method to interact with a machine by voice will be described with reference to the drawings.

FIG. 4 is a functional block diagram showing the configuration of a conventional voice recognition response device. As shown in the figure, the AD converter 1 converts the input voice into a digital signal. The acoustic analysis unit 2 analyzes the voice for each analysis section (frame) and obtains spectrum information (linear prediction coefficient). The characteristic parameter extraction unit 3 obtains the LPC Ceptrum coefficient, the normalized logarithmic residual power C _{0, the} band pass filter value, etc. using the linear prediction coefficient. The frame sync signal generator 4 generates a sync signal for each frame. When performing pattern matching, the standard pattern selection unit 9 sequentially selects standard patterns of recognition target words and phrases. The partition candidate setting unit 6 sets a plurality of candidate voice section lengths for each selected standard pattern. The time-axis normalization unit 5 expands / contracts the time-series pattern of the characteristic parameters for the set plurality of candidate voice sections to a fixed frame length. The similarity calculation unit 7 calculates the similarity with each standard pattern in the standard pattern storage unit 8 selected by the standard pattern selection unit 9. The similarity comparison unit 11 compares the maximum similarity values stored in the primary storage 10 for each frame, and selects the recognition target phrase having the maximum similarity. The word having the maximum similarity obtained in the primary memory 10 is updated. The words and phrases remaining in the primary storage 10 are output to the task control unit 12 from the similarity comparison unit 11 as a recognition result. The task control unit 12 selects a response sentence voice corresponding to the recognition result, and gives a command to the voice synthesis unit 13. The voice synthesis unit 13 takes out voice data of a specified sentence from the voice data storage unit 14 and synthesizes voice. The synthesized digital voice is converted into a DA converter.
Converted to analog at 15 and output.

The task control unit 12 outputs a recognition command to the similarity comparison unit 11,
The recognition is started to obtain the recognition result, and the speech synthesis unit 13
By repeating the reproduction of the voice of the response sentence with the synthesis command for, the flow of the entire target task is controlled.

Problems to be Solved by the Invention However, in the configuration of the conventional example as described above, FIG.
As shown in (d), pattern matching is performed also for a voice that has started to be uttered before the start of recognition, that is, a so-called truncated voice, and misidentification of selecting the most likely candidate from the candidates, or Since the likelihood is low, it may not be recognized (reject). If it is determined to be rejected, a synthetic voice or a signal sound that prompts a re-voice is issued, and the user must receive the same and re-speak the same phrase or sentence. Further, in the case of erroneous recognition, it is necessary to instruct the device to recognize it and generate a word or the like to be recognized again. For this reason, there is a problem in that a smooth dialogue process is not performed, which is inconvenient for the user.

The purpose of the present invention is to judge whether the voice is truncated or recognizable at the start of recognition, and in the case of the loss of head, a speaker who notices an error spontaneously without interrupting the flow of processing by reject. (EN) A voice recognition responding device capable of continuously re-speaking, recognizing the voice, and performing a smooth and natural dialogue.

Means for Solving the Problems The present invention achieves the above object, and comprises means for detecting the presence or absence of an input voice and voice recognition processing means, and at the time of starting the voice recognition by the recognition processing means, When a voice is detected by the detection means, the recognition processing is stopped, and the recognition processing is started for the input voice re-spoken after the silent section is detected by the detection means. is there.

Operation The present invention has the above-described configuration, at the start of recognition, it is determined whether or not the voice is uttered, and if it is uttered, the recognition process is not started, and the recognition process is started only after the silent section is detected. . If the speaker makes a mistake in his or her own utterance, it immediately utters it again. It is possible to recognize and process the correct input voice.

Example Hereinafter, an example of the present invention will be described in detail with reference to the drawings.

First, a first embodiment of the present invention will be described. FIG. 1 is a functional block diagram showing a configuration of a voice recognition response device in a first embodiment of the present invention.

In the figure, each part indicated by reference numerals 1 to 15 has the same configuration as that of the conventional example described in FIG.
Only different configurations will be described.

The sound detecting unit 17 compares the power value of the bandpass filter obtained by the characteristic parameter extracting unit 3 with a certain threshold value and detects a frame exceeding the threshold value. When the recognition result generation unit 16 detects that the filter power value exceeds the threshold value by the sound detection unit 17 at the time of starting recognition in response to the recognition command from the task control unit 12, the recognition process is immediately performed. Do not start. After that, the silence detecting unit 19 detects a frame in which the power value of the bandpass filter is equal to or less than a certain threshold value, that is, a silent section. The recognition result generation unit 16 starts the recognition process after the silent section is detected.

Generally, a speaker, if he or she notices an error in his or her own utterance, immediately utters it again. , The correct input voice immediately re-speaked by the speaker can be subsequently recognized and processed.

FIGS. 2A to 2E are views showing the states of the above processing. As shown in the figure, the input voice uttered before the start of recognition is not recognized, and the re-voiced voice is recognized.

Next, a second embodiment of the present invention will be described. FIG. 3 is a functional block diagram showing the configuration of the second embodiment of the present invention.

In the above-described first embodiment, since it is determined whether or not the speech has a truncated word at the start of recognition, there is a risk that recognition may be erroneously stopped for instantaneous noise other than speech. In addition, there is a risk that the recognition process is erroneously started by detecting a silent sound in a voice or a silent unvoiced consonant portion having a low power while the recognition is stopped. Therefore, in the present embodiment, consideration is given to erroneously stopping the recognition or not erroneously starting the recognition process with respect to instantaneous noise and audible sounds.

In the figure, each part indicated by reference numerals 1 to 17 and 19 is the same as the configuration of the first embodiment shown in FIG. 1, so the description thereof will be omitted and only different configurations will be described. The voiced frame counting unit 18 counts the number of frames in which the bandpass filter power value output from the voice detecting unit 17 exceeds a certain threshold from the recognition start command time, and when a voiced frame of a certain fixed frame or more continues And outputs a recognition stop request to the recognition result generation unit 16 and the silent frame counting unit 20. The silence frame counting unit 20 detects a silence frame output from the silence detecting unit 19 after a recognition stop request, detects a time point at which a certain number of frames or more continue, and issues a recognition process restart request to the recognition result generation unit 16
Output to.

After that, the same processing as in the first embodiment is performed.

In each of the above embodiments, the presence or absence of voice is detected by the bandpass filter power, but the same detection can be performed by using the entire range power.

EFFECTS OF THE INVENTION As described above, according to the present invention, when voice recognition is detected by the means for detecting the presence or absence of an input voice at the start of voice recognition by the voice recognition processing means, the recognition processing is stopped. The recognition process is started for the input voice re-speaked after the silent section is detected by the detection means. That is, the presence / absence of voice is judged at the start of recognition, and if no voice is input yet, the recognition process is performed thereafter. After that, the recognition process is performed on the words uttered after the silent section. Therefore, if the speaker notices an erroneous utterance and spontaneously utters it again,
Without interrupting the flow of the dialogue between the person and the device due to misrecognition or rejection due to a head loss, the correct input voice that has been uttered again is continuously recognized and processed, so that a smooth and natural dialogue can be realized.

[Brief description of drawings]

FIG. 1 is a functional block diagram showing the configuration of a voice recognition response device in a first embodiment of the present invention, FIG. 2 is a timing diagram showing a state of dialogue in an embodiment of the present invention, and FIG. 2 is a functional block diagram showing the configuration of the second embodiment, FIG. 4 is a functional block diagram showing the configuration of a conventional voice recognition response device,
FIG. 5 is a timing chart showing a state of dialogue in the conventional example. 1 ... AD conversion unit, 2 ... Acoustic analysis unit, 3 ... Feature parameter extraction unit, 4 ... Frame synchronization signal generation unit, 5 ... Time axis normalization unit, 6 ... Section candidate setting unit, 7 ... ... Similarity calculation unit, 8 ... Standard pattern storage unit, 9 ... Standard pattern selection unit, 10 ... Primary storage, 11 ... Similarity comparison unit, 12 ...
Task control unit, 13 ... voice synthesis unit, 14 ... voice data storage unit, 15 DA conversion unit, 16 ... recognition result generation unit, 17 ...
Sound detection section, 18 ... Sound frame counting section, 19 ... Silence detection section, 20 ... Silence frame counting section.

Claims (3)

[Claims] 1. A means for detecting the presence / absence of an input voice and a voice recognition processing means. When the voice recognition is started by the recognition processing means, when the voice is detected by the detection means, the recognition is performed. A voice recognition response device, characterized in that the processing is stopped, and a recognition process is started for an input voice re-spoken after a silent section is detected by the detection means. 2. A sound detector for comparing a power value of an audio signal with a certain threshold and detecting a frame exceeding a threshold, and a silence detector for detecting a frame below the threshold. The voice recognition response device according to claim 1, which is configured. 3. A sound detector for comparing a power value of an audio signal with a certain threshold and detecting a frame exceeding a threshold, and a silence detector for detecting a frame below the threshold. The voice recognition response device according to claim 1, comprising a voiced frame counting unit and a silent frame counting unit for calculating the number of frames output from each detection unit.