gustapoll/tts-bororo

MMS Bororo TTS (LoRA)

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

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=320):

Metric	In-space	Out-of-space
CER	0.467	0.462
MCD	844.4	851.5

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

Try it live

gustapoll/demo — a Gradio Space with both ASR and TTS tabs for Bororo, 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-bor")

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