-
https://git.ecker.tech/ aims to provide a place to share my efforts while maintaining true ownership of my code, as I do not trust GitHub.
XMR: 4B9TQdkAkBFYrbj5ztvTx89e5LpucPeTSPzemCihdDi9EBnx7btn8RDNZTBz2zihWsjMnDkzn5As1LU6gLv3KQy8BLsZ8SG
- Joined on
2022-10-10
Do you have a loss target in mind?
Not necessarily a target loss, but moreso a mix of playing it by ear from the output, and the reported AR accuracy being >90%. I can't remember what loss…
Fixed in commit 805d7d35e8436d0684ae9bf568a03c2f945fadd9.
Added in commit baa6b76b85bdbd04296a134bd1116d60c622908e. Refer to the Use via API thing at the bottom of the web UI page. It should automatically handle reloading if TTS is already loaded.
Actual weekly progress report:
I feel very, very stupid for burning so much time being stubborn. Restarting the weights was actually the correct call, as the results are looking pretty good.…
Alrighty, it should be added in commit 31da215c5f0f77a249dd5131bf8c3a4f149299a3 (and mrq/tortoise-tts commit 5ff00bf3bfa97e2c8e9f166b920273f83ac9d8f0).
So, to reiterate:
- Two types of…
31da215c5f
added checkboxes to use the original method for calculating latents (ignores the voice chunk field)
5ff00bf3bf
added flags to rever to default method of latent generation (separately for the AR and Diffusion latents, as some voices don't play nicely with the chunk-for-all method)
5ff00bf3bf
added flags to rever to default method of latent generation (separately for the AR and Diffusion latents, as some voices don't play nicely with the chunk-for-all method)
9221c66d03
added checkboxes to use the original method for calculating latents (ignores the voice chunk field)
056b6f2f91
added flags to rever to default method of latent generation (separately for the AR and Diffusion latents, as some voices don't play nicely with the chunk-for-all method)
056b6f2f91
added flags to rever to default method of latent generation (separately for the AR and Diffusion latents, as some voices don't play nicely with the chunk-for-all method)
If you're looking to simply utilize the web UI's transcription routine for subtitles, I suggest just using whisper (or whisperX) directly…
mmm. I suppose I'll add in a way to use the original, vanilla latents computation.
I did just remember, that maybe about a month or so after modifying the latents computation to use the entire…
I might have a "guide" (list) somewhere in an Issues reply (although I don't think I did this, it's not something I can just give a simple guide on and expect it to cover all grounds), but the…
Hmm...
- check under
Settings, I think it'sOutput Sample Rate, check if it's set to 44100 (I think I might have left it default to 44000, but it shouldn't be the cause). You can also set…
Right, I keep forgetting to re-implement a CLI interface. When I get a moment I can whip up something.
First, the train.py script doesn't use anything from TorToiSe directly, so I'm not sure what you're doing to get that from the given information.
Secondly, the intended fork to use it with…