JPH04156600A - Voice recognizing device - 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] Industrial applications The present invention relates to a speech recognition device.

BACKGROUND OF THE INVENTION Conventionally, acoustic information including voice is collected by an acoustic collecting section,
In a speech recognition device that performs speech recognition by sending a signal of acoustic information collected by the acoustic collection section to a speech recognition section, it is extremely difficult to recognize speech uttered in noise. This is not only due to the problem of noise being added to the voice input to the speech recognition device, but also due to the deformation of the utterance itself due to the Lombard effect, which was shown by Lombard in 1911. The Lombard effect occurs because when speaking in a noisy environment, the speaker tries to speak louder and clearer because it becomes difficult to hear his or her own voice.

Problems to be Solved by the Invention As described above, in speech recognition under noisy conditions, the problem of noise being added to speech has been widely discussed. As a countermeasure against this problem, the well-known adaptive noise canceling method is being considered. In addition, countermeasures against the Lombard effect are available in Proceedings of the Acoustical Society of Japan ■, l-3-9, pp.
, 17-18, March 1990, etc., have been considered, but methods that make the Lombard effect less likely to occur have not been considered much. Further, in Japanese Patent Application Laid-open No. 61-221928, a method is considered in which the speaker is always made to hear noise so as to make the speaker speak loudly, but no method has been considered to make the Lombard effect less likely to occur.

Means for Solving the Problem Therefore, in order to solve such problems, claim 1
In the described invention, in a speech recognition device that performs speech recognition by collecting acoustic information including speech by an acoustic collecting section and sending a signal of the acoustic information collected by the acoustic collecting section to a speech recognizing section, A high-efficiency audio collection unit is provided in the unit to collect audio with a good signal-to-noise ratio even in a high-noise surrounding environment, and the audio input to the audio collection unit equipped with the high-efficiency audio collection unit is fed back to the speaker. A feedback department was established.

In the invention as claimed in claim 2, a noise estimating section is provided for estimating surrounding noise, and a noise removing section is provided for removing noise from input speech using the noise estimated by the noise estimating section.

The invention according to claim 1 can reduce the reduction in recognition rate due to the Lombard effect, since the sound input by the sound collecting section for speech recognition is fed back to the speaker. It is possible to provide a speech recognition device that can reduce speech fluctuations caused by the Lombard effect and that can be recognized even under high noise conditions.

The invention set forth in claim 2 further combines the invention set forth in claim 1 with a speech recognition device and a noise removal device, so that it is possible to further reduce noise components other than speech, and thereby, it is possible to further reduce noise components other than speech. It is possible to realize a speech recognition device that can also be used in

EXAMPLE An example of the invention as claimed in claim 1 will be described based on FIG. As shown in FIG. 1, acoustic information including speech is collected by an acoustic collecting section 1, and a speech signal a of the acoustic information collected by this acoustic collecting section 1 is sent to a speech recognition section 2 to perform speech recognition. In the speech recognition device that performs
The sound collecting section 1 is provided with a highly efficient sound collecting means (not shown). This high-efficiency audio collection means has the function of collecting audio with a high signal-to-noise ratio even in a highly noisy surrounding environment.

Further, an audio feedback unit 3 is connected to the audio collecting unit l equipped with such a highly efficient audio collecting means. This audio feedback unit 3 includes the audio collecting unit 1
The function is to feed back the input voice to the speaker.

In this case, the sound collecting section 1 can collect the sound uttered by the speaker using, for example, a microphone. Note that it is desirable to use a close-talk type microphone or a directional microphone as the microphone.

A close-talking microphone is close to the speaker's mouth, so it is possible to input voice with a good signal-to-noise ratio. A directional microphone can input voice with a good signal-to-noise ratio by matching the direction of its direction to the mouth of the speaker. However, if you input audio using something other than such a microphone, the SN ratio of the collected audio will be poor, and as a result, the audio to be fed back will also be affected.
Care must be taken because a lot of surrounding noise may be mixed in, making it difficult for the speaker to understand his/her own voice.

Further, as the audio feedback section 3, for example,
Using headphones worn by the speaker or speakers placed near the speaker, the collected voice can be fed back to the speaker to make it easier to hear the speaker's own voice.

Further, the speech recognition unit 2 can be implemented using, for example, a known speech recognition system.

In such a configuration, acoustic information including voice and surrounding environment noise other than the voice is collected by the acoustic collecting section 1. The high-efficiency audio collecting means provided in the acoustic collecting unit 1 can collect audio with a high SN ratio even in a high noise environment, so the audio signal a collected in this way can be
is sent to the speech recognition section 2, where speech recognition is performed.

In addition, at this time, since the voice signal a collected with a good S/N ratio is fed back to the speaker via the audio feedback unit 3, it is possible to reduce the decrease in the recognition rate due to the Lombard effect. You can make it easier to open and understand your own voice.

As described above, since the audio signal a obtained by the high-efficiency audio collection means of the audio collection unit 1 is fed back to the speaker by the audio feedback unit 3, it is possible to reduce audio fluctuations due to the Lombard effect under high noise conditions. Moreover, this makes it possible to provide a speech recognition device that can recognize speech even under high noise conditions.

Next, an embodiment of the invention according to claim 2 will be described based on FIG. Here, the following configuration is added to the configuration described in the embodiment of the invention described in claim 1 (see FIG. 1).

That is, when the surrounding environment is noisy, the speech recognition device according to the embodiment of the invention described in claim 1 has the problem that a certain amount of noise is added to the speech signal, and the recognition rate may decrease depending on the case. There is also. Therefore, in this embodiment, a noise estimation section 4 for estimating the surrounding noise is provided before the speech recognition section 2, and the noise estimated by the noise estimation section 4 is used to remove noise from the input speech. A noise removing section 5 is provided to remove the noise.

In such a configuration, the audio signal a collected by the acoustic collecting section 1 is input to the noise removing section 5. At this time,
The noise estimating unit 4 estimates the noise of the surrounding environment based on the input signal from the acoustic collecting unit 1 in the section where the audio signal a does not exist, thereby obtaining an estimated noise amount. For example, the average value of the spectral pattern of the noise input over a certain period of time is determined. Then, the estimated noise amount estimated in this manner is sent to the noise removal section 5, whereby the estimated noise amount is removed from the input audio signal a. Thereafter, the speech signal a from which noise has been removed in this manner is sent to the speech recognition section 2, where speech recognition is performed.

As mentioned above, since the speech recognition device is combined with the noise removal device including the noise estimation unit 4 and the noise removal unit 5, it is possible to further reduce noise components other than speech. This makes it possible to realize a speech recognition device that can be used even under high noise conditions.

Next, a modification of the above-mentioned invention according to claim 2 will be explained based on FIG. 3. Here, the noise estimating section 4 and the noise removing section 5 are not used, and in addition to using a voice inputting micro as an acoustic collecting section for voice input, a noise inputting microphone 7 is separately provided as an acoustic collecting section for noise input. Furthermore, an adaptive noise canceling method is employed in which an adaptive filter 8 is connected to the noise input microphone 7.

In this case, the noise component is adaptively removed by inputting the noise signal obtained from the noise input microphone 7 to the audio signal a obtained from the audio input micro via the adaptive filter 8. This makes it possible to obtain an audio signal, thereby achieving the same effects as in the embodiment described above (see FIG. 2).

Effects of the Invention The invention according to claim 1 provides a voice recognition system that performs voice recognition by collecting acoustic information including voice by an acoustic collecting section and sending a signal of the acoustic information collected by the acoustic collecting section to a speech recognition section. In the apparatus, the sound collecting section is provided with a high-efficiency sound collecting means that collects sound with a good signal-to-noise ratio even in a high noise environment, and the sound input to the sound collecting section equipped with the high-efficiency sound collecting means is uttered. Since an audio feedback unit is provided to provide feedback to the speaker, the voice input by the acoustic collecting unit for speech recognition is fed back to the speaker, which can reduce the decrease in recognition rate due to the Lombard effect. Accordingly, it is possible to provide a speech recognition device that can reduce speech fluctuations due to the Lombard effect under high noise conditions and can recognize speech even under high noise conditions.

The invention according to claim 2 is provided with a noise estimating section that estimates surrounding noise, and a noise removing section that removes noise from input speech using the noise estimated by the noise estimating section. Since the invention described in 1 is further combined with a speech recognition device and a noise removal device, it is possible to further reduce noise components other than speech, and thus it can be used even in high noise environments. This makes it possible to realize a speech recognition device.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing an embodiment of the invention as claimed in claim 1, Fig. 2 is a block diagram showing an embodiment of the invention as claimed in claim 2, and Fig. 3 is a block diagram showing an embodiment of the invention as claimed in claim 2. It is a block diagram showing a modification. 1... Sound collection section, 2... Speech recognition section, 3... Voice feedback section, 4... Noise estimation section, 5... Noise removal section

Claims (1)

[Claims] 1. A speech recognition device that performs speech recognition by collecting acoustic information including speech by an acoustic collecting section and sending a signal of the acoustic information collected by the acoustic collecting section to a speech recognizing section, The sound collecting section is provided with a high efficiency sound collecting means that collects sound with a good signal-to-noise ratio even in a high noise surrounding environment, and the sound input to the sound collecting section equipped with the high efficiency sound collecting means is fed back to the speaker. A voice recognition device characterized by being provided with a voice feedback section that provides voice feedback. 2. The apparatus according to claim 1, further comprising a noise estimating section for estimating surrounding noise, and a noise removing section for removing noise from input speech using the noise estimated by the noise estimating section. Speech recognition device.