Research And Realization On Timbre Conversion Technique For Tibetan Amdo Dialect

The present paper have accomplished three research objectives, namely analysed timbre ofTibetan Amdo dialect, framed sound model of Tibetan Amdo dialect and realized the timbreconversion system of Tibetan Amdo dialect.To achieve above research objective, this papercompleted the following study, which based on text and voice.1. The aim of this study is to design a Tibetan Amdo dialect corpus with facing timbreconvert and establish its basic research. In order to start follow-up study of this paper , inthe first place , the paper have to solve one question, that is how to design a corpus whichbased on timbre convert. It is compulsory that the corpus can cover differentcombination of consonant and vowel on Tibetan Amdo dialect, this corpus also be able torepresent rhythm features recording corresponding sound corpus and segmenting,marking sound file after design text corpus.2. The analysis of acoustic features of time-domain and frequency, timbre reflects speakers'personality traits, which reflected rhythm features on acoustic characteristics, they arepitch, length, energy and pause, also reflected spectrum features, namely spectralcentroid, spectral slope and formant frequency. By analyzing different acoustics soundfeature, we can find key trait which can reflect timbre conversion for Tibetan Amdodialect.3. Established Tibetan Amdo dialect timbre model by analysed acoustics characteristicsparameters and its extraction. Each Tibetan Amdo word represents one syllable, but itsvowel sound system is quite different with Chinese word, and its syllable without tone, itis necessary to full examine Tibetan Amdo dialect syllable structure under timbreconversion algorithm for Tibetan Amdo,and structuring mode according tocharacteristics of syllable.4. Achieved timbre conversion system for Tibetan Amdo dialect . In the system ofimplementation process, STRAIGHT algorithm is the key , we can accomplish timbreconversion system for Tibetan Amdo by using of high-quality speech analysis/synthesisalgorithm STRAIGHT, and improved this algorithm., accomplished process designdescription and system performance analysis. I evaluated computational complexity andusing the subjective evaluation method (such as ABX test, MOS test, etc.) to evaluatethe results of covert.