http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
A study on Fuzzy Vector Quantizer Mapping in Speech Synthesis for Speaker Adaptation
이광형,Lee, Jin Yi,You, Jae Tack 대한전자공학회 1994 ISPACS:Intelligent Signal Processing and Communica Vol.1 No.1
In this paper, we propose a speaker-adaptive speech synthesis method using a mapped codebook designed by fuzzy mapping Competitive learning neural networks are used to design both input speaker's and reference speaker's codebook. We used a fuzzy VQ mapping to design the mapped codebook. The fuzzy VQ mapping replaces a codevector preserving its fuzzy membership function. The codevector correspondence histogram rs obtained from accumulating Use vector correspondence along Use DTW optimal path. Using each histogram as a weighting function, the mapped codebook is defined as a linear combination of the reference speaker's vectors. Speech synthesis is performed as follows. input speaker's speech is fuzzy vector quantized by the mapped codebook, and then FCM arithmetic is used to synthesize the speech adapted to input speaker. The speaker adaption experiments are carried out using speech of male in his thirties, speech of a male in his twenties, and speech of a female in her twenties, as input speaker's speech, and speech of a other female in her twenties, as reference speaker's speech. Speech used in experiments are sentences / anyoung hasim nika / and / good morning / As a results of experiments, we have listened a synthesized speech adapted to input speaker.
A Study on the Speech Recognition using Pitch Synchronous LPC Cepstrum-VQ and HMM
이광형,고남곤,Lee, Jin Yi,Kim, Hyung Seuk 대한전자공학회 1994 ISPACS:Intelligent Signal Processing and Communica Vol.1 No.1
In this paper we propose a phoneme recognition model of the Korean speech by using the pitch synchronous LPC cepstrum-VQ and FIMM. The LPC cepstrum coefficient have been obtained by synchronizing the analysis frame per a pitch. The pitch synchronization method reduces the analysis trine and the effect of the pitch pulses. The LPC cepstrum of the Input speedy signal is vector quantized by using the reference codebook of the LPC cepstrum coefficient. This codebook has already been designed with the FCM clustering algorithm. The LPC cepstrum codevector's address in codbook is applied to the f IMM algorithm, and then recognized phoneme by phoneme. As the speech signal vary slowly with tune, the HMM technique is, in general, effective in it modeling in time. The performance evaluation of, the proposed method has been done for both inside training data and outside training data. As the simulation results, it can be shown that the proposed recognition method has improved about 3(%) in recognition ratio than the pitch asynchronous method.