remove redundant phonemize for vall-e (oops), quantize all files and then phonemize all files for cope optimization, load alignment model once instead of for every transcription (speedup with whisperx)

oops

cleanups, realigning vall-e training

VALL-E config edits

added japanese tokenizer (experimental)

cleanup, metrics are grabbed for vall-e trainer

forgot to separate phonemes by spaces for [redacted]

cleaned up some prepare dataset code

unk hunting

oops

preparations for training an IPA-based finetune

added options to pick tokenizer json and diffusion model (so I don't have to add it in later when I get bored and add in diffusion training)

removed redundant training data (they exist within tortoise itself anyways), added utility: view tokenized text

;)

(:

:)

added mel LR weight (as I finally understand when to adjust the text), added text validation on dataset creation

cleanups and fixes, fix DLAS throwing errors from '''too short of sound files''' by just culling them during transcription

only God knows why the YAML spec lets you specify string values without quotes

added the mysterious tortoise_compat flag mentioned in DLAS repo

forgot template

big cleanup to make my life easier when i add more parameters

actually make using adamw_zero optimizer for multi-gpus work

disable validation if validation dataset not found, clamp validation batch size to validation dataset size instead of simply reusing batch size, switch to adamw_zero optimizier when training with multi-gpus (because the yaml comment said to and I think it might be why I'm absolutely having garbage luck training this japanese dataset)

made validation working (will document later)

set validation to save rate and validation file if exists (need to test later)

added option to set worker size in training config generator (because the default is overkill), for whisper transcriptions, load a specialized language model if it exists (for now, only english), output transcription to web UI when done transcribing

renamed mega batch factor to an actual real term: gradient accumulation factor, fixed halting training not actually killing the training process and freeing up resources, some logic cleanup for gradient accumulation (so many brain worms and wrong assumptions from testing on low batch sizes) (read the training section in the wiki for more details)

added new training tunable: loss_text_ce_loss weight, added option to specify source model in case you want to finetune a finetuned model (for example, train a Japanese finetune on a large dataset, then finetune for a specific voice, need to truly validate if it produces usable output), some bug fixes that came up for some reason now and not earlier

huge success

Added very experimental float16 training for cards with not enough VRAM (10GiB and below, maybe) \!NOTE\! this is VERY EXPERIMETNAL, I have zero free time to validate it right now, I'll do it later

added more safeties and parameters to training yaml generator, I think I tested it extensively enough

oops

added dropdown to select autoregressive model for TTS, fixed a bug where the settings saveer constantly fires I hate gradio so much why are dropdown.change broken to contiuously fire and send an empty array

a bit of UI cleanup, import multiple audio files at once, actually shows progress when importing voices, hides audio metadata / latents if no generated settings are detected, preparing datasets shows its progress, saving a training YAML shows a message when done, training now works within the web UI, training output shows to web UI, provided notebook is cleaned up and uses a venv, etc.

almost