Read a loud integration with Piper neural TTS

[anzo42]

Hi, I came across a cool project on Github https://github.com/rhasspy/piper that lets you use local neural voices for text-to-speech conversion that outperforms any SAPI/synthesized speech system I came across so far.
The project runs on Linux/Mac/PC systems, so it fits perfectly into Thorium's distribution scheme, it has a fairly extensive base of supported languages with many voice options, plus the author points out that the engine is optimized to run on Raspberry, so it should be fairly light on system resources, too.

I recommend at least exploring what kind of voice quality this engine can offer on this page https://rhasspy.github.io/piper-samples/ and perhaps assessing whether this is something Thorium can benefit from

[danielweck]

Thank you for sharing this.

Under the hood, Thorium interfaces with the standard Web Speech API exposed by the Chromium web browser engine. For example on Windows there are SAPI5 bindings, so TTS voices compatible with this protocol are automatically picked-up by Thorium.

Nowadays there is a wide choice of free and commercial LLM / neural network powered text-to-speech libraries, but most seem to require ad-hoc interfacing, either via online / HTTP requests or some kind of offline binding API.

Thorium's internals would need to be refactored in order to add support for TTS beyond the Web Speech API. This is not something we have the resources to work on at the moment, I am afraid.

I am aware of third-party efforts to bridge SAPI5 in Windows with online Azure voices, OneCore etc. (mentioned in another thread in this discussion forum), so I imagine this could be technically feasible with other technologies?

[panaC]

Interesting, thank you

[danielweck]

https://github.com/Lej77/windows-text-to-speech?tab=readme-ov-file#text-to-speech-on-windows

[anzo42]

Thanks for explaining how things works and what are the limitations in this regard.
Also a big thanks for the link to the Lej77 project, while I was able to use this method for Windows Narrator without any problems, unfortunately, with Thorium itself, after switching the built-in TTS to Lej77's voice, it did not work for me. The sentence to be read was outlined, but the narration did not start, and the individual words were not highlighted.
Nevertheless, I will try to dig a little more on the matter.