Research On Speech Endpoint Detection Algorithm With Low SNR

Speech endpoint detection of speech signal refers to finding out the beginning and ending point of signal accurately.That is to say,we can separate speech signal from noise background.The purpose is providing guarantee for the follow process.When the signal-to-noise ratio(SNR)is high,the endpoint detection that the traditional two-threshold endpoint detection method,the method based on spectral entropy and variance endpoint detection method can be good to complete the task of detection and have a good accuracy.However,in the process of voice collection,the SNR of the speech signal becomes gradually lower due to different voice collection environments and different types of noise,and the conventional detection method cannot be better to complete the detection task.So it is very important to study the endpoint detection method when the signal to noise ratio is relatively low.In the paper,the noisy speech signal is processed though the improved spectral subtraction based on multitaper spectral estimation.It can achieve the purpose of reducing noise.Then speech signal with reducing noise is detected by using the method of BARK subband variance in frequency domain.By detecting the speech signal with-5dB Gaussian noise,the experimental results are obviously better than those of other algorithms.On this basis,we can change the way to improve the resolution and combine the improved spectral subtraction based on multitaper spectral estimation again.We can verify the feasibility of this method.When the speech signals with white noise,pink noise,volvo noise and f16 cockpit noise added to 20 dB to-20 dB are detected by endpoint detection,the accuracy is also relatively well compared with other methods.