Request access to mos-mms-tts

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

mos-mms-tts is a GO AI Corp model shared for non-commercial research (CC BY-NC 4.0). Commercial use needs written permission from contact@goaicorporation.org. Please read the card, its terms and licenses before requesting access, and make sure you understand them.

mos-mms-tts

Mooré text-to-speech: MMS-TTS (VITS) fine-tuned by GO AI Corp.


Task	Text-to-speech, Mooré (`mos`)
Base model	MMS-TTS (VITS); the starting checkpoint was not recorded, most likely facebook/mms-tts-mos
Training data	Not recorded
Published	July 2024 (updated August 2024)
Related	Served by the product endpoint mos-tts

Résumé en français. Synthèse vocale en mooré : MMS-TTS (VITS) affiné par GO AI Corp. Le point de départ n'a pas été noté (très probablement facebook/mms-tts-mos), pas plus que les données d'entraînement. Publié en juillet 2024 (mis à jour en août 2024) ; c'est le modèle servi par le point d'accès mos-tts. Licence CC BY-NC 4.0 ; tout usage commercial demande l'autorisation écrite de contact@goaicorporation.org. Accès sur demande.

Quick start

import torch, scipy.io.wavfile
from transformers import VitsModel, AutoTokenizer

name = "goaicorp/mos-mms-tts"
model, tok = VitsModel.from_pretrained(name), AutoTokenizer.from_pretrained(name)
with torch.no_grad():
    wav = model(**tok("Ne y yibeoogo.", return_tensors="pt")).waveform[0].numpy()
scipy.io.wavfile.write("out.wav", model.config.sampling_rate, wav)

Evaluation

Test set	CER (ASR loopback)	UTMOS	Speaker similarity
Mooré held-out set (to be fixed)	pending	pending	pending

CER is measured by transcribing the synthetic speech with a fixed Mooré ASR model.

Limits

Training data is not documented.
One voice; no voice cloning.

License

CC BY-NC 4.0. Commercial use needs written permission from contact@goaicorporation.org.

Citation

@misc{goaicorp2024mosmmstts,
  title        = {mos-mms-tts: Mooré text-to-speech with MMS-TTS},
  author       = {Nyantudre, Wendyellé Abubakrh Alban and Bandaogo, Aristide},
  year         = {2024},
  howpublished = {Hugging Face Models, GO AI Corp},
  url          = {https://huggingface.co/goaicorp/mos-mms-tts}
}

Developed and maintained by

Wendyellé A. Alban NYANTUDRE: nyantudre@gmail.com
Aristide BANDAOGO: aristide@goaicorporation.org

Downloads last month: 8

Safetensors

Model size

36.3M params

Tensor type

F32

Model tree for goaicorp/mos-mms-tts

Base model

facebook/mms-tts-mos

Finetuned

(2)

this model

Finetunes

1 model

goaicorp
/

mos-mms-tts