-
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
Oct 10, 2022
mrq commented on issue mrq/ai-voice-cloning#427
Google Colab Notebook Not WorkingI can't remember the exact incident, but the colab hasn't been supported in a while now, but: > `OSError: /usr/local/lib/python3.10/dist-packages/torchaudio/lib/libtorchaudio.so: undefined…
mrq commented on issue mrq/ai-voice-cloning#426
Kickstart foreign language training using XTTS weights?Pretty much how you outlined it. You'll just need to prepend every line under `train.txt` with `[pl]`, and make sure to supply your modified tokenizer and cleaner code inside DLAS [here](https:/…
mrq pushed to master at mrq/vall-e
- ed54f4ebec un 'experimental' the better target sequence preparation
mrq pushed to master at mrq/vall-e
- 9a6040383e make validation samplers ignore sampler type
mrq pushed to master at mrq/vall-e
- 32d4271ca8 fixed issue with training from scratch (oops)
mrq commented on issue mrq/ai-voice-cloning#421
Finetuning diffusion modelIt's a mixed bag. [152334H/DL-Art-School](https://github.com/152334H/DL-Art-School#training-the-diffusion-model-wip) *should* have diffusion finetuning covered. If I remember right, it should…
mrq commented on issue mrq/ai-voice-cloning#418
Two trainings with exact same parameters result in different curves> but I was wondering if that was normal? For model training, it *can* boil down to the initial seed, but in the scope of finetuning TorToiSe's AR model, I don't think it's worth the trouble.…
mrq commented on issue mrq/tortoise-tts#49
Learning rate finderIt's a bit of a pickle, since all the relevant code to actually train the model is within [DLAS](https://git.ecker.tech/mrq/DL-Art-School). I believe [here](https://git.ecker.tech/mrq/DL-Art-School…
mrq commented on issue mrq/ai-voice-cloning#416
Can't get the model training started> I liked one of the random generated voices, how do I use it? If you happened to have `Embed Output Metadata` enabled, you can take the output with the random latents you want, and drag and…
mrq pushed to master at mrq/vall-e
- 3195026dba fixed issue with the 'add another target audio to artificially create longer sequences' for HDF5 just duplicating the utterance initially sampled
mrq commented on issue mrq/ai-voice-cloning#417
Line delimiter is heard in the output> EDIT - After many tests, I am pretty sure the artifacts are created by the [I am really {emotion},] at the start of the line, may it be manually written or implied by the selection of one of the…
mrq pushed to master at mrq/vall-e
- 09cda7d3f9 added sampling by speaker group name (might be better to de-emphasize the LibriVox/Audiobooks that are in large numbers, and emphasize the smaller pools), log cleanup
mrq pushed to master at mrq/vall-e
- 82329703d3 cull speakers with too few utterances from the group sampler
mrq pushed to master at mrq/vall-e
- 69e746711a added sampling by speaker group name (might be better to de-emphasize the LibriVox/Audiobooks that are in large numbers, and emphasize the smaller pools), log cleanup
mrq commented on issue mrq/ai-voice-cloning#416
Can't get the model training started> [Training] [2023-10-15T14:45:21.813802] local_state[k] = v[grad_accum_step] > [Training] [2023-10-15T14:45:21.813802] IndexError: list index out of range Your gradient accumulation size…
mrq commented on issue mrq/ai-voice-cloning#415
RESTful API?Either through gradio's exposed API that's shown at the bottom of the web UI's page, or through [the WebSocket server](https://git.ecker.tech/mrq/ai-voice-cloning/src/branch/master/src/api/websocke…
mrq pushed to master at mrq/vall-e
- a539f6889f mucked around with the loss calculation, this seems better?
mrq pushed to master at mrq/vall-e
- fb467b19ba exposed rolling resp context to the web UI, added passing in language to inferencing command line
mrq pushed to master at mrq/vall-e
- 65f500083d tweaks to try and get deepspeed quantized inferencing, validating bitsandbytes and deepspeed quantization, nothing seems to work