mms_finetune_kbd_murat

MMS-TTS Fine-tuned for Kabardian (Speaker: Sokhov Murat)

This repository contains a fine-tuned version of Facebook's MMS-TTS model, adapted for generating speech in the Kabardian language. The model is trained on a dataset of audio recordings by the speaker Sokhov Murat.

Model Details

• Base Model: facebook/mms-tts
• Fine-tuned on: anzorq/kbd_speech dataset
• Training steps: 5,100
• Speaker: Sokhov Murat
• Language: Circassian (Kabardian)

Usage

To use this model for text-to-speech generation, you can leverage the pipeline functionality from the Transformers library. Here's an example:

from transformers import pipeline
import scipy

model_id = "anzorq/mms_finetune_kbd_murat"
synthesiser = pipeline("text-to-speech", model_id, device=0) # add device=0 if you want to use a GPU

text = "дауэ ущыт?"
speech = synthesiser(text)

# Save the generated audio to a file
scipy.io.wavfile.write("finetuned_output.wav", rate=speech["sampling_rate"], data=speech["audio"][0])

This code will generate an audio file finetuned_output.wav containing the speech synthesis for the provided Kabardian text.

Notes

• Fine-tuned following the guide at https://github.com/ylacombe/finetune-hf-vits
• Since no pre-trained MMS-TTS model was available for Kabardian, we fine-tuned a model for Chechen, which has the closest character set to Kabardian.
• Do not use in production. This model's performance is considerably worse than that of the fine-tuned VITS model anzorq/kbd-vits-tts-male for Kabardian text-to-speech.

License

The original MMS-TTS model by Meta is licensed under the CC-BY-NC-4.0 License. This fine-tuned version inherits the same license.

Acknowledgments

• AI at Meta for the original MMS-TTS model.
• Sokhov Murat for providing the audio recordings used for fine-tuning.