From 6bf496b9edb95504290a0388c3c9d0f73bf0388c Mon Sep 17 00:00:00 2001 From: mrq Date: Wed, 26 Apr 2023 04:53:55 +0000 Subject: [PATCH] Update Generate --- Generate.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/Generate.md b/Generate.md index 124f25b..f51b67c 100644 --- a/Generate.md +++ b/Generate.md @@ -102,9 +102,11 @@ You can set `autoregressive_model` to `auto` alongside a `voice` to automaticall ### Additional Backends -Some additional backends have not-as-tight integration. You can specify a backend by passing `--tts-backend="backend name"` when launching the starting script. +Some additional backends have not-as-tight integration. You can specify a backend by passing `--tts-backend="backend name"` when launching the starting script. These backends are ***highly*** experimental, so a lot of features that work for TorToiSe are unavailable for these backends. I tried my best to hide what's not needed, but I'm not perfect. * `vall-e`: Utilizes [mrq/vall-e](https://git.ecker.tech/mrq/vall-e) as the TTS backend + Requires the repo to be cloned to `./modules/vall-e/` and installed using `pip3 install -e ./modules/vall-e/` * `bark`: Utilizes [suno-ai/bark/](https://github.com/suno-ai/bark/) as the TTS backend - + Requires the repo to be cloned to `./modules/bark/` and installed using `pip3 install -e ./modules/bark/` \ No newline at end of file + + Requires the repo to be cloned to `./modules/bark/` and installed using `pip3 install -e ./modules/bark/` + + Requires the target voice to have already been transcribed using the `Training > Prepare Dataset` tab. + - This is because creating custom speaker prompts require a transcription of the reference prompt, and what better way to do that than leverage existing transcriptions. \ No newline at end of file