Hi
i'm Tarannom.
i would like to have matlab code for endpoint detection in speech recognition, i was searching and i saw your page.
i'm leaving in iran and i am studing master of science in ALZAHRA university of Tehran.
i need Matlab code of endpoint detection based on zaro crossing rate and short term energy
thanks alot
i really need your help.
Posts: 14,118
Threads: 61
Joined: Oct 2014
Speech processing designates a team composed of Prof. Lawrence Rabiner (Rutgers University and University of California, Santa Barbara), Prof. Ronald Schafer (Stanford University), Kirty Vedula and Siva Yedithi (Rutgers University). This exercise is part of a set of speech processing exercises intended to complement the didactic material in the textbook "Theory and applications of digital speech processing" by L R Rabiner and R W Schafer.
This MATLAB exercise analyzes an audio file that contains a period of background signal, followed by a spoken speech period, and ending with a background signal period. The analysis is performed on a frame-by-frame basis and, based on a set of short-time timing velocity parameters and zero timing, the program determines the frame that is the best estimate of the beginning of the speech AND the frame Is the best estimate of the end of the voice signal.
The exact location of the ends of an isolated word is important for reliable and robust word recognition. The endpoint detection problem is not trivial for non-stationary backgrounds in which artifacts (ie non-speaking events) can be introduced by the speaker, the recording environment, and the transmission system. Several techniques were used to detect the extremes of isolated words recorded through a dialed telephone line. The techniques were broadly classified as explicit, implicit or hybrid in the concept. Explicit techniques for the detection of endpoints locate the endpoints before and independently of the recognition and decision stages of the system. For implicit methods, the endpoints are determined only by the recognition and decision steps of the system, ie there is no separate step for endpoint detection. Hybrid techniques incorporate aspects of both explicit and implicit methods. Researches showed that hybrid techniques consistently provided the best estimates for the two extremes of words and, consequently, the greater accuracy of recognition of the three classes studied. A hybrid endpoint detector is proposed which gives a rejection rate of less than 0.5 percent, while providing near-accurate recognition accuracy from manually modified endpoints.
