Voice signal endpoint detection method based on characteristic value code

Abstract

The invention discloses a voice signal endpoint detection method based on a characteristic value code. Characteristic parameters including a short-time energy and a short-time zero-crossing rate are extracted in frames, an average value of the short-time energy, an average value of the short-time zero-crossing rate and a maximal value of the short-time zero-crossing rate are calculated, four thresholds are set for the short-time energy via a statistical result and empirical parameters, one threshold is set for the short-time zero-crossing rate, a voice characteristic is coded according to the thresholds, and endpoint detection is carried out on a voice signal according to five-grade determination rules on the basis of the characteristic value code of each frame. The lowest threshold is set for the short-time energy of an audio segment, suspected voices are accepted / rejected according to the rules by combining characteristics of the adjacent frames, the five-grade determination rules can handle with different complex conditions effectively, false detection and neglected detection are avoided, and the correct rate of endpoint detection of the voice signal can be improved substantially.

Description

technical field [0001] The invention belongs to the field of speech signal processing, and in particular relates to a speech signal endpoint detection method based on eigenvalue coding. Background technique [0002] Speech signals can be divided into voiced segments and silent segments, and voiced segments can also be consonant segments, vowel segments, and transition segments between consonants and vowels. In speech recognition and speaker recognition systems, the mixing of silent segments will significantly reduce the recognition performance of the system, and the start and end positions of each vocal segment must be detected. This is the endpoint detection technology of speech signals. [0003] The speech signal has short-term stable characteristics. The short-term characteristics of the speech signal are extracted through frame processing. The short-term characteristics of the time domain mainly include short-term energy and short-term zero-crossing rate. Endpoint detect...

AI Technical Summary

Problems solved by technology

[0010] In practice, it has been found that when using the traditional double-threshold method to detect consonants, as long as the short-term energy is lower than EL and the short-term zero-crossing rate is higher than ZH, it is judged as a consonant, ignoring the minimum requirement for short-term energy in voiced segments, often Cause false detection or missed detection

In addition, it is also found that some consonants have a more obvious initial segment and stronger energy, but when the tail is close to the vowel, the energy is weakened and the zero-crossing rate is also reduced.

[0011] In addition, the result of speech signal endpoint detection is heavily dependent on the thresholds of short-term energy and short-term zero-crossing rate, and how to set these thresholds, there is no unified and recognized method. If the threshold is not selected properly, it is easy to cause false detection. Or missed detection, the correct rate of endpoint detection is significantly reduced

Images