Text to Speech, Voice Synthesis, Re: Can it be done?

Question

So, I'm quite new here and I had difficulty in searching for some information.

I'm wondering if this technology currently exists on a mobile platform.

I want to use a phone to record a persons voice and phrase(s). Then based upon the voice, allow another person to type in a text to retrive the voice data. Preferably not using the same words, while at the same time giving a phrase to capture the correct syllables needed to produce the desired phrase. Basically text to speech, but real time input, kind of like from Mission Impossible 2. E.G my brother phones me, I record his voice, then phone my mom and sound like my brother. However rather than piece the voice together like in music files, type in what I want to say. Perferably something that doesn't sound mechanical.

Hopefully my description makes sense and any help would be greatly appriciated.

Thanks, Will

Answer 1

This functionality you are looking for is called "voice conversion". The current technology can adapt one person voice to sound like another person. It's enough to record about 30 sentences of target voice to do the adaptation. You don't presumably need to synthesize from the text, it's easier to convert voice from other person's voice.

There are demos of this technology, for example check Festival:

http://festvox.org/transform/transform.html

For the commercial providers of this technology you can check

http://clonevoice.com/en

TTS with voice adaptation is also available in open source tools, but it's more complex technology than voice transform, see Speaker Adaptive Training here:

http://hts.sp.nitech.ac.jp/?Download

Answer 2

Text to speech engines use a database of prerecorded voice audio where there is a clip for every possible syllable. Unless you get your brother to record every syllable you wouldn't be able to synthesize his voice.