On 13/11/25 03:30, Samuel Thibault wrote:
online TTS will always be a problem, better focus on packaging offline models which can be downloaded once for good, and kept loaded into memory so there isn't the delay of loading it.
I fully concur. As an aside (not necessarily an example of the above), I found the following additional multilingual model recently:
https://github.com/resemble-ai/chatterbox The speech quality of the online demonstration is impressive.