Research on Automatic Generation Algorithm of Phoneme Conversion Learning Corpus Based on KNN Algorithm

As for phoneme conversion, there is no good method all the time, so most of the current researches on speaker voice conversion only convert the phonetic spectrum. From the linguistic point of view, phonemes are the smallest unit of speech, and syllables and sentences are all composed of phonemes in a fixed way. In this paper, the research on automatic generation algorithm of phoneme conversion learning corpus based on KNN(K nearest neighbor) algorithm is carried out. In order to solve the problem that it is difficult to collect a large amount of speech data of the target user, this paper uses KNN algorithm to construct an average phoneme model, mixes the data in the previously established phoneme database, and obtains GMM (Gaussian Mixture Model) of phonemes through sufficient training. In 10 cases, the spectral distance ratio based on KNN is lower than that based on GMM, with an average reduction of 9.261%, which indicates that independent conversion of each phoneme by phoneme classification can effectively avoid GMM smoothing and obtain more information of the target speaker.