Synthesize speech waveform from HMMs trained by HTS
