JPS62111291A - Noise level learning apparatus - 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 Application Field) The present invention detects a speech section from an input sound in which a speech section and a section containing only surrounding noise without speech are continuous in speech recognition, etc. The present invention relates to a noise level learning device for ambient noise.

(Prior Art) FIG. 3 is a functional block diagram of a conventional noise level learning device. In FIG. 3, 31 is the handset of the car telephone, 32 is the preprocessing section, 33 is the lock release lever that unlocks the handset 31, and 34 is the release of the lock release lever 33 (corresponding to the off-hook of a normal telephone). , 36 is a noise level learning section, and 37 is a subsequent section such as voice section detection and voice recognition. This conventional example is an example of a noise level learning device for detecting voice sections in a voice recognition device for a car phone.

Next, the operation of the above conventional example will be explained with reference to FIG.

When the speaker unlocks the handset to pick it up,
Detect it (■) and learn the noise level (■)
. To learn the noise level, the power level of the noise is observed for 0.4 seconds after the cancellation of the noise is detected, and the noise level is learned as the average level during that time.Then, based on the result, speech section detection, speech recognition, etc. Do it (■).

In this way, even with the conventional noise level learning device described above, noise level learning is possible to some extent.

(Problems to be solved by the invention) However, in the above conventional noise level learning device,
There were the following problems. Namely.

Car phone handsets are often installed between the driver's and passenger's seats, but this location has a high level of noise such as gear noises, and the noise level between the time the handset is unlocked and the time when the handset is actually lifted is high. Moreover, if the handset is picked up and noise is generated, it will be learned as part of the surrounding noise. As long as a method is used to detect a voice section, if a continuous input with a level higher than the level of the surrounding noise is detected as a voice section, it is not possible to learn a level that is higher than the original level during noise learning.

This causes difficulty in detecting voice sections.

The present invention solves these conventional problems,
It is an object of the present invention to provide a noise level learning device that can perform noise level learning at the correct level.

(Means for Solving the Problems) In order to achieve the above object, the present invention provides an audio input means (11, 12) for inputting audio, and a noise input device for learning a noise level based on the output of the audio input means. Level learning means (16) and means for receiving a speaker's utterance preview (14)
) and means (15) for causing the noise level learning means to start learning the noise level after a predetermined period of time has passed after receiving the one-voiced preview.

(Function) Therefore, according to the present invention, noise level learning is performed after a certain period of time has elapsed after the speaker's utterance notice.
Even if a large level of noise is input to the voice input means immediately after the voice announcement, it can be excluded from the noise level learning target, and the effect is that highly accurate noise level learning can be performed.

(Example) FIG. 1 shows an embodiment of the present invention.

Like the conventional example, this embodiment is also an example of a voice recognition device for a car telephone. In FIG. 1, 11 is a handset which is a means for inputting voice, 12 is a front part, a series part, 13
．． Reference numeral 14 indicates a means for receiving a voice announcement from a speaker, 13 indicates a lever for unlocking the handset, and 14 indicates a lever 13.
15 is a section that delays the detection signal of the detection section 14, 16 is a noise level learning section, and 17 is a subsequent section such as voice section detection and voice recognition.

Next, the operation of the above embodiment will be explained with reference to FIG. When the speaker unlocks the handset 11 to pick it up, the unlock detection unit 14 detects this (■
). The detection signal is delayed for 0.3 seconds by the delay unit 15 (■). After a delay of a certain period of time (0.3 seconds), the noise level learning section 16 starts learning the noise level (■). To learn the noise level, the power level of the noise is measured for 0.4 seconds, and the noise level is learned as the average level during that time. After that,
The post-processing unit 17 performs post-processing such as voice section detection and voice recognition based on the results.

In this way, according to this embodiment, the learning of the noise level is
By not doing this for a certain period of time after unlocking the handset, it is possible to prevent learning of the gear sounds mentioned in the explanation of the conventional example and the loud noises when lifting the handset, and to improve accuracy. It has the advantage of being able to learn noise levels.

(Effects of the Invention) As is clear from the above embodiments, the present invention enables highly accurate noise level learning by performing noise level learning after a certain period of time has elapsed after one speaker's utterance notice has been received. This has the advantage that, as a result, the accuracy of speech segment detection and speech recognition performed later is also improved.

[Brief explanation of drawings]

Fig. 1 is a functional block diagram of an embodiment of the present invention, Fig. 2 is an explanatory diagram of the operation of the same embodiment, Fig. 3 is a functional block diagram of a conventional noise level learning device, and Fig. 4 is an explanatory diagram of its operation. It is. 11.31...Handset, 12.32...Preprocessing unit. 13.33...Lever to unlock the handset,
14.34...Handset unlock detection unit, 15.
...Delay part of the same detection signal. 16.36... Noise level learning section, 17.37...
- Post-processing unit for voice section detection, voice recognition, etc. Patent applicant: Matsushita Electric Industrial Co., Ltd. Figure 1 Figure 2

Claims (3)

[Claims] (1) A voice input means for inputting voice; a noise level learning means for learning a noise level based on the output of the voice input means; a means for receiving a voice preview of a speaker; A noise level learning device comprising means for causing the noise level learning means to start learning the noise level after a period of time has elapsed. (2) The noise level learning device according to claim (1), which uses a telephone handset as the voice input means. (3) The noise level learning device according to claim (2), wherein the telephone is a car telephone, and the means for receiving the speaker's utterance notice is a lever for unlocking the handset.