Broadband background noise and voice separation detection system and method

Abstract

The invention relates to the field of information processing technology and sensing signal processing and more particularly to a broadband background noise and voice separation detection system. The system comprises a current frame time-domain energy calculation circuit, a background noise calculation circuit, a time-domain voice detection long-and-short-time average energy comparison circuit, a frequency-domain voice detection length time-frequency-domain energy comparison circuit, a background noise comparison circuit, a sub-band energy distribution uniformity voice detection circuit, and a voice frame number counting circuit. The invention also discloses a broadband background noise and voice separation detection method. The system and the method use a three-level voice detection means, has a very good detection effect on high-low-frequency background noise, and has a very good detection effect on the noise of the accidental intermittent line, and greatly improves the accuracy of voice detection in the complicated noise environment.

Description

technical field [0001] The invention relates to the fields of information processing technology and sensor signal processing, in particular to a wide-band background noise and speech separation detection system and method. Background technique [0002] A hot spot in the field of artificial intelligence applications is speech recognition, which has been widely used in various fields. The implementation of speech detection is an important part of the real-time realization of the speech recognition system. Its purpose is to distinguish the speech segment from the non-speech segment in a complex actual environment. Some literature shows that the low recognition rate in practical applications is largely due to the lack of recognition of the speech segment. Correct processing, a large amount of non-speech noise seriously affects the accuracy of the speech recognition system, especially for speech recognition with a lot of noise in the application environment, the correct speech de...

AI Technical Summary

Problems solved by technology

[0003] The current mainstream method of automatic voice endpoint detection is to rely on the short-term energy in the time domain, the zero-crossing rate, and the mean square error of the frequency band energy in the frequency domain. The specific method is to find out the short-term energy, zero-crossing rate or The mean square error of frequency band energy is then compared with an empirical threshold. Experiments have proved that this method of comparing short-term energy or zero-crossing rate alone is not suitable for noisy environments, especially when the application environment changes. The same When the background noise of the environment will also change, the frequency band energy mean square error method is not suitable for quiet environments

[0004] Speech detection can also be performed separately according to the change of the average sound energy in the time domain and frequency domain, and finally the optimal result can be selected according to the dynamically estimated background noise, thereby greatly improving the accuracy of speech recognition and the adaptability to environmental changes , since the energy of most steady-state background noise is concentrated in the low-frequency band, this method is very effective for most low-frequency distribution noises, and for sounds made by objects or animals such as birdsong, car horns, pianos and other musical instruments Sound, due to its wide frequency band distribution, is within the range of the voice band distribution of the same person. For this type of noise, it is easy to misjudge this type of noise as speech if the above method is used. Distinguishing this type of noise is useful for speech detection and speech noise reduction. , speech recognition is very important, but also one of the difficulties

Images