Run own ipa to speech model ?

[camarm-dev]

Currently, to retreive the word pronunciation in audio file, Remède use a third party. Make search to run our own model right in the API.

Comments can contains ideas, articles, repos links ect...

[camarm-dev]

ipa-reader.xyz (our current provider) seems to use classic TTS (https://cuttlesoft.com/blog/2018/09/13/pronouncing-things-with-amazons-polly/)

[camarm-dev]

Juste need to find a TTS model that:

• Support french
• Support SSML tag <phoneme>
• Is open sources

Links to projects in thread.

[camarm-dev]

Is it worth to use pronunciation available on wikipedia ?

• Scrap all pronunciations using another script and index them in a database (ipa -> file path)
• If no pronunciation, generate using Amazon Polly (not open source 😢)
  Can be worth but it don't seems to be stable...

[camarm-dev]

Using https://github.com/gmn/nanotts for the moment. It does not support phoneme SSML tag.
Converting this issue and mark this disscussion as resolved.

We should look for new TTS projects which can help us.