Mms Spa Finetuned Chilean Monospeaker

🚀 Transformers Text-to-Speech Model

This project provides a text-to-speech solution using the finetuned MMS model. It can generate high - quality Spanish speech with low latency, trained on a Chilean Spanish dataset.

✨ Features

• Light - weight and Low - latency: Based on the VITS architecture, it offers efficient text - to - speech conversion.
• Fast Training: Can be finetuned in around 20 minutes with as few as 80 to 150 samples.
• Multi - platform Support: Usable in both Python (Transformers library) and JavaScript (Transformers.js).

📦 Installation

Transformers.js

If you haven't already, you can install the Transformers.js JavaScript library from NPM using:

npm i @xenova/transformers

💻 Usage Examples

Basic Usage - Python (Transformers)

from transformers import pipeline
import scipy

model_id = "ylacombe/mms-spa-finetuned-chilean-monospeaker"
synthesiser = pipeline("text-to-speech", model_id) # add device=0 if you want to use a GPU

speech = synthesiser("Hola, ¿cómo estás hoy?")

scipy.io.wavfile.write("finetuned_output.wav", rate=speech["sampling_rate"], data=speech["audio"])

Advanced Usage - JavaScript (Transformers.js)

import { pipeline } from '@xenova/transformers';

// Create a text-to-speech pipeline
const synthesizer = await pipeline('text-to-speech', 'ylacombe/mms-spa-finetuned-chilean-monospeaker', {
    quantized: false, // Remove this line to use the quantized version (default)
});

// Generate speech
const output = await synthesizer('Hola, ¿cómo estás hoy?');
console.log(output);
// {
//   audio: Float32Array(69888) [ ... ],
//   sampling_rate: 16000
// }

Optionally, save the audio to a wav file (Node.js):

import wavefile from 'wavefile';
import fs from 'fs';

const wav = new wavefile.WaveFile();
wav.fromScratch(1, output.sampling_rate, '32f', output.audio);
fs.writeFileSync('out.wav', wav.toBuffer());

📚 Documentation

Model

This is a finetuned version of the Spanish version of Massively Multilingual Speech (MMS) models, which are light - weight, low - latency TTS models based on the VITS architecture.

It was trained in around 20 minutes with as little as 80 to 150 samples, on this Chilean Spanish dataset.

Training recipe available in this github repository: ylacombe/finetune - hf - vits.

📄 License

This model is licensed under the cc - by - nc - 4.0 license.