PLP及MFCC在躲语连续语音识别系统中的比拟(国外英文资料).docVIP

PLP及MFCC在藏语连续语音识别系统中的比较(国外英文资料) PLP and MFCC are compared in a continuous speech recognition system in Tibetan 【 abstract 】 this paper discusses the common speech feature parameters, and analyzes the MEL frequency cepstrum coefficient (MFCC) and perception and the calculation method of linear prediction coefficient (PLP) and big words in the Tibetan capital Lhasa words table continuous speech recognition system of extraction of MFCC and PLP parameters respectively, and the recognition results are compared. [keywords] MEL inverted spectral perception linear prediction coefficient MFCC PLP speech recognition Tibetan Automatic speech recognition research began in the 1950 s and 80 s biggest breakthrough is the application of hidden markov model (HMM), speech recognition research focus from speaker-dependent, list of small, isolated word speech recognition to speaker-independent, big words, continuous speech recognition was removed; Since the 90 s, the speech recognition in the models, parameter extraction and optimization, the system made breakthrough adaptive. Entering this century, the famous research institutes and companies, such as the university of Cambridge, university, Microsoft, IBM, cmu bell laboratories of institutions such as big words table of continuous speech recognition system on a particular speaker recognition rate reached 95%. In the face of the future market in China, foreign companies such as IBM, apple, MOTOROLA into the development of Chinese speech recognition system. Speech recognition research in China started late, but development is rapid, automation research institute, institute of acoustics, Chinese academy of sciences and tsinghua university, Beijing jiaotong university and other institutions have carried out the research of speech recognition, on the whole, Chinese continuous speech recognition research and foreign advanced technology. Actual environmental requirements for acoustic noise robustness of speech recognition is higher and hi