Extraction Of Tibetan Syllable Based On Feature Recognition

Speech recognition in the present is the research hotspot, speech recognition is involved in an interdisciplinary subject to many disciplines of linguistics, computer science, signal processing, physiology, psychology, and artificial intelligence in the field of pattern recognition is the important branch. However, Tibetan speech recognition research is lagged behind. This paper mainly introduces the Tibetan monosyllabic preprocessing and feature extraction method.The preprocessing plays a vital role in speech recognition.. For Tibetan speech recognition pretreatment study reference is English monosyllabic speech processing method. In the speech recognition system in the correct decision, starting point and end point of speech input for improving the recognition rate is very important. Obvious limitations for transient signal, non stationary process, a signal with broadband noise signal using the traditional method of processing, in high signal-to-noise ratio, noise filtering is very easy, using short-time energy magnitude and short time average zero crossing rate can get better detection results. First summarize the advantages of the cepstral parameters, cepstral parameters can not only reflect the characteristics of the original signal and for harmonic and noise almost can obviously distinction in, has good robustness and stability, so of speech recognition, now almost all the cepstrum. In the speech recognition system, the use of the spectrum is more advantageous. This article uses several methods of comprehensive treatment can accurate rate higher recognition speech endpoint, and in low signal to noise ratio of noise filtering, recognition of unvoiced sounds and voiced sounds. Experiments show that this method is correct and improves the accuracy of detection.Feature extraction is to retain as much as possible to identify the effective information, while minimizing the maximum of what useless, redundant information. Because the technology of speech processing is applied in many aspects, and in the speech recognition system, the key is the extraction of the feature parameters of speech.. When speech recognition, the final recognition results are related to the preprocessing, and the feature parameters of the signal are more important, which directly determines the results of the identification.. Speech signal feature extraction is very important in speech signal processing.. This paper begins to study the basic knowledge, mainly in speech signal processing, speech recognition technology, and the most basic principle. For LPCC, the feature parameters have a lot of relationship with background noise, because the linear prediction of the frequency of the noise is very sensitive, and the Mel frequency coefficients is not affected by the noise.. For this article, we have to study and deal with the noise containing speech signal for the purpose of the paper.. Because in a short time interval, the time domain features of the speech signal can be remained basically unchanged, which is to study speech signal foundation can have the results, this paper uses Mel frequency cepstrum coefficient(MFCC) feature extraction of the monosyllabic Tibetan.