gustapoll/tts-paresi

MMS Paresi TTS (LoRA)

A LoRA adapter (r=8, alpha=16, targeting q_proj/v_proj) fine-tuning facebook/mms-tts-fin for text-to-speech (TTS) in Paresi, a Brazilian indigenous language. Trained with a combined CTC + cycle-consistency (TTS resynthesis) + mel-reconstruction loss as part of the ablation variant no_pseudo.

Note: Paresi is not in MMS's native language set. This adapter was bootstrapped from MMS's fin checkpoint (facebook/mms-tts-fin) as the closest available base, then fine-tuned on Paresi data.

Non-commercial use only

The base MMS model is licensed CC-BY-NC-4.0. This adapter inherits that restriction: non-commercial use only.

Quality caveat

This is an early-stage, low-resource fine-tune, evaluated on a small held-out set (n=30):

Metric	In-space	Out-of-space
CER	0.345	0.306
MCD	990.0	1041.7

Treat outputs as a research demo, not production-quality transcription/synthesis.

Try it live

gustapoll/paresi-demo — a Gradio Space with both ASR and TTS tabs for Paresi, also callable as an API via gradio_client.

How to use

from transformers import VitsModel, AutoTokenizer
from peft import PeftModel

base = VitsModel.from_pretrained("facebook/mms-tts-fin")

model = PeftModel.from_pretrained(base, "REPO_ID").eval()
tokenizer = AutoTokenizer.from_pretrained("REPO_ID")