Speech breakpoint detection (short time improved subband spectral entropy)

1. Audio analysis

1. Output speech segmentation time point information , The time point is expressed in milliseconds ;
2. Split the speech into multiple wav file ;

Endpoint detection ： Determine the time starting point and ending point of the sentence , Ignore a small number of non voice frames in the middle ,
For speech recognition .(Speech Endpoint Detection)

Entropy is a quantity that reflects information measurement in information theory . The greater the randomness of a random event ,
That is, the higher the uncertainty , The greater the entropy , So the more information you carry .
This operation adopts Spectral entropy method End point detection for voice .

2. Spectral entropy method

3. Preprocessing

4. Double threshold method endpoint detection

5. experimental result

Handle PHONE_001.wav Information obtained
（1）time.csv: Segment information for voice ;
（2）PHONE_001_vad.wav: For voice VAD After processing , Speech segment synthetic wav;
（3）segmentation Folder : It is the speech of each segment after speech segmentation ;
（4）main_VAD.m: The main function ;
（5）vad.m: It is the endpoint detection function of double threshold method ;
（6）houzhichuli.m： Is the interval length decision function ;
（7）frame2time.m： As a function of time for a frame ;