custom pronunciation

[sankar-mukherjee]

This is char based tts. does it support custom pronunciation?

[psdwizzard]

I know for the screen reader app that I built as a Chrome extension, I have a tab in there for custom dictionaries. For example, John=Jon (like if the H was throwing off the tts system.) that. And I'm thinking about implementing that in my custom fork, but I haven't gotten a chance to do it yet. It's on the roadmap, though.

[AznamirWoW]

This is char based tts. does it support custom pronunciation?

I've found that the TTS does pronounce things quite differently based on the reference audio. With a sample of David Attenborough's voice the output sounds like a british gentleman, with a sample of Donald Trump the output sounds like an american dumbass.

[psdwizzard]

@AznamirWoW While you're 100% right about accents, I have noticed that at least for XTTS2, all accents for the most part will pronounce certain things wrong, which is why in my Chrome extension I have a custom dictionary of pronunciations.

[psdwizzard]

[AznamirWoW]

@AznamirWoW While you're 100% right about accents, I have noticed that at least for XTTS2, all accents for the most part will pronounce certain things wrong, which is why in my Chrome extension I have a custom dictionary of pronunciations.

XTTS v2 has very good English, but the pronunciation does not really change when you change the speaker. Other languages perform somewhere between good and terrible.

Chatterbox has noticeable differences as said. Example: chatterbox.zip

[MercyfulKing]

This is char based tts. does it support custom pronunciation?

I've found that the TTS does pronounce things quite differently based on the reference audio. With a sample of David Attenborough's voice the output sounds like a british gentleman, with a sample of Donald Trump the output sounds like an american dumbass.

So the output is accurate, nice.

[tim-basic]

I've noticed in my own tests on the huggingface space that cloning with UK English accent works well. In my tests, I used Stephen Fry reading Harry Potter and it came out well. However, when using audio samples of Australian news reporters or samples of my own voice (I have Aus accent) for cloning, it came out with a strong UK English accent.

Are there ways to set parameters or tweak things to get Aus accent / different English accents?

[tim-basic]

Actually, now I've just done aus accent tests with audio from here and it worked very well ... hmm

[tim-basic]

hmm ... I've done some more tests using audio from here. It's a bit hit and miss. Seems to perform worse with female voices. It matches pitch and timbre okay but accent is lost.