Font Size: a A A

The Speech Endpoint Detection In Highly Noisy Background And The Realization On DSP

Posted on:2008-09-14Degree:MasterType:Thesis
Country:ChinaCandidate:T HanFull Text:PDF
GTID:2178360242965037Subject:Circuits and Systems
Abstract/Summary:PDF Full Text Request
Along with the technical development of the multi-media technique and the communications technique, the voice signal processing as the basic contents, have already been applied in different area. With the development of the technique, the dynamic range of the signal handled become wider and wider. But in the actually practical, because of the influence of background noise, inaccuracy voice activity detection produced many unnecessary processing times and lowered the accuracy of the voice signal processing. So various voice activity detection algorithm have applied in different voice signal processing system.Separate the speech and the background noise on the time domain called the voice endpoint detection. Nowadays speech endpoint was an important segment in the speech preprocess. It is has become an essential component in the communications system, speech coding, speech enhancement, voice recognition, echo cancer. The previous methods were based on the short time energy and zero crossing rate, MFCC, pitch detection. But the performance of these methods becomes poor in the lower SNR.This paper summary the traditional voice endpoint detection algorithm, and get into the heart of this algorithm's basic principle, compare each advantage and disadvantage and give the simulation result. Analyzed the strut of the speech signal and the characteristic parameter and give two new algorithms. The one is based on the multi-band entropy and the other one is the algorithm based on the Gaussian-Gamma model. And then give the simulation result. The simulation experiment demonstrated that the algorithm had a better robustness. These algorithms improved the precision and robustness of the endpoint detection. It had a good performance even in low SNR environments.And then, this paper has carried on feasibility research on how to implement the algorithm on the hardware platform. The core of the hardware processing platform is the TI Company's high performance DSP of the TMS320DM642; TLVAIC23B is the speech input and output chip. Design the soft and hardware system of the voice endpoint detection system. And describe the software process and code optimization. The experiments show these algorithms has high reliability makes it suitable for application in different kinds of environments.At the end of the paper, the two proposed algorithms are summarized. Some new study fields within the past two years are introduced and developing perspective of the voice activity detection is referred to.
Keywords/Search Tags:Voice Endpoint Detection, Gaussian-Gamma model, Entropy, Digital Signal Process
PDF Full Text Request
Related items