Umbaji/eyaa-tom-mms-tts-kdh

MMS-TTS Kotokoli + Tem — Eyaa-Tom Fine-tuned

Fine-tuned version of facebook/mms-tts-kdh on the Eyaa-Tom dataset for Kotokoli + Tem (kdh).

Kotokoli and Tem are the same language (ISO kdh). Trained on merged data from both folders.

Language Details

Field	Value
Language	Kotokoli + Tem
ISO 639-3 (MMS)	kdh
Your ISO	kdh / kot
Region	Togo/Ghana
Family	Gur (Niger-Congo)
Base model	facebook/mms-tts-kdh

Usage

from transformers import VitsModel, VitsTokenizer
import torch, torchaudio

model     = VitsModel.from_pretrained("Umbaji/eyaa-tom-mms-tts-kdh")
tokenizer = VitsTokenizer.from_pretrained("Umbaji/eyaa-tom-mms-tts-kdh")

inputs = tokenizer("your text here", return_tensors="pt")
with torch.no_grad():
    waveform = model(**inputs).waveform[0]

torchaudio.save("output.wav", waveform.unsqueeze(0), model.config.sampling_rate)

Training Details

• Loss: Mel-spectrogram L1 (avoids VITS training restriction)
• Optimizer: AdamW — lr=2e-4, betas=(0.8, 0.99)
• Scheduler: ExponentialLR γ=0.999
• Epochs: 6 | Batch size: 4 (effective 16 w/ grad accumulation)

Citation

@article{pratap2023mms,
  title={Scaling Speech Technology to 1,000+ Languages},
  author={Pratap, Vineel et al.},
  journal={arXiv preprint arXiv:2305.13516},
  year={2023}
}

Fine-tuned: 2026-02-25 — Eyaa-Tom project