Having the computer speak is key to human-like communication. However, the set of voices available for text-to-speech system (TTS) is limited. In this workshop we will be building a voice for the open source FestVox/Festival TTS that you record. The computer talks like you! (Or whomever you convince to record the prompts for you – your dog?).
The workshop is also a hands-on introduction to the free book “Building Synthetic Voices” that comes with Festival TTS.
You don’t need programming knowledge but you need some Unix knowledge. If you want to extend the recipes we will see in the workshop you will need to learn Scheme which is the programming language for extending Festival.
The workshop presentations will be in English, but we are more than happy to provide support and answer questions in French. Moreover, we will only cover generating English voices, as we are following the default tooling provided by FestVox. We hope to cover French synthesis in a later workshop.
We will cover:
- How to install and set-up Festival/FestVox/Edinburgh Speech Tools
- Overview of the full Festival TTS stack
- How to design prompts for limited domain TTS (reduced vocabulary)
- A full example in the bicycle safety domain that you’ll record during the workshop. (Hear it synthesized with Pablo’s voice here.)
- Introduction to diphone (large vocabulary) recording and synthesis (no diphone recording will take place during the workshop, though)
This workshop is just the beginning, potential follow-ups include French synthesis, differentiated intonation recordings or any other topic that interests the group. If you have a project that uses speech synthesis (or might benefit from it) bring a demo and we can hear it at the end of the workshop.
Requirements:
- A computer with a unix-like environment (GNU/Linux, Mac OS X or cygwin) capable of doing sound recording.
- RSVP to pablo.duboue (at) gmail.com required. Space is limited.
- Cost: $10 (all proceedings from this workshop will be donated to Foulab)
When?
Saturday, January 24th at 15h (about 3 hours in length)
Where?
Foulab, of course!
