* RESEARCH TOPICS - SPEECH SYNTHESIS [#m6b8da40] //&ref(speechsynthesis.jpg,nolink,center,40%); ** Introduction [#o2a165e8] > The speech synthesis is technology that let a computer do natural utterance like the human being.~ We push forward a study to make more fluent and natural utterance. ** HMM Speech Synthesis [#u1f78200] > Various technique such as the unit choice type is suggested to a voice synthesis, but we treat technique based on a statistical frame such as Hidden Markov model (HMM) mainly in our laboratory.~ The HMM speech synthesis system is based on HMM, and this models quantity of characteristic to be included in a sound statistically and synthesize a sound. ** Research Field [#k18cf145] *** Text Speech Synthesis [#z66e5bde] > This is technology that synthesizes a sound from an arbitrary text.~ This is called Text to Speech (TTS) too.~ [[Demonstration page(Japanese page)>http://www.sp.nitech.ac.jp/~demo/gtalk/demo.php]] [[Demonstration page (Japanese page)>http://www.sp.nitech.ac.jp/~demo/gtalk/demo.php]] *** Speaker Interpolation [#d3c8bafc] > This is technology that creates the models of a new speaker by putting the model of plural speakers together.~ [[Demonstration page>HOME/DEMONSTRATION/SPEAKER INTERPOLATION]] *** Eigen Voice [#o28cbc54] > This is technology that expresses plural speakers and styles of the speech in model space by a little parameter and synthesizes the sound of arbitrary quality of voice by adding heaviness to those parameters.~ [[Demonstration page>HOME/DEMONSTRATION/EIGEN VOICE]] *** Singing Voice Synthesis [#x08960e4] > This is technology that synthesizes an arbitrary song by giving a text and score.~ We can expect various uses in the field of entertainment and the amusement.~ [[Demonstration page>HOME/DEMONSTRATION/SINGING VOICE SYNTHESIS]] *** Speech Convertion [#d7027006] > This is technology that converts the sound of a certain speaker into as if the sound of another speaker.~ This learns models from the utterance data of two speakers and converts the utterance data of one speaker into the data of other speaker.~ *** Speaker Adaptation [#g8d70d7f] > This is technology that converts the quality of voice of the aim speaker by converting a model parameter of HMM from a little data of the speaker of the aim.~ [[Demonstration page>HOME/DEMONSTRATION/SPEAKER ADAPTATION]]