mohamedrayyan/mms-tts-div-ft-spk01-m01

MMS-TTS Dhivehi — fine-tuned checkpoint

Dhivehi text-to-speech checkpoint fine-tuned from facebook/mms-tts-div (VITS architecture). Trained as part of the Dhivehi TTS project (mohamedrayyan/chatterbox-tts-dhivehi).

Usage

from transformers import VitsModel, AutoTokenizer
import torch

model = VitsModel.from_pretrained("mohamedrayyan/mms-tts-dhivehi-ft-spk01-m01")
tokenizer = AutoTokenizer.from_pretrained("mohamedrayyan/mms-tts-dhivehi-ft-spk01-m01")

text = "މިއީ ދިވެހި ބަހުން އަޑުލައިދޭ ނަމޫނާއެކެވެ."
inputs = tokenizer(text, return_tensors="pt")
with torch.no_grad():
    audio = model(**inputs).waveform

Variants

Repo	Description
mohamedrayyan/mms-tts-dhivehi-ft-spk01-m01	female voice 01
mohamedrayyan/mms-tts-dhivehi-md-f02	female voice 02
mohamedrayyan/mms-tts-dhivehi-md-f03	female voice 03
mohamedrayyan/mms-tts-dhivehi-md-m01	male voice 01
mohamedrayyan/mms-tts-dhivehi-spk01-f01	voice-cloned speaker 01 (f)
mohamedrayyan/mms-tts-dhivehi-spk01-m01	voice-cloned speaker 01 (m)

Quantized (mms-tts-dhivehi-quantized) and ONNX (mms-tts-dhivehi-onnx) exports are also published. Sample audio in samples_mms/ of the project repo.