MMS-TTS-YOR: Yoruba Text-to-Speech Model
| Property | Value |
|---|---|
| Developer | Facebook (Meta AI) |
| License | CC-BY-NC 4.0 |
| Model Type | VITS (Variational Inference with adversarial learning for TTS) |
| Paper | Scaling Speech Technology to 1,000+ Languages (2023) |
What is mms-tts-yor?
MMS-TTS-YOR is a specialized text-to-speech model designed for the Yoruba language, developed as part of Facebook's Massively Multilingual Speech (MMS) project. This model utilizes the VITS architecture to provide end-to-end speech synthesis capabilities, converting Yoruba text into natural-sounding speech.
Implementation Details
The model implements a conditional variational autoencoder (VAE) architecture with three main components: a posterior encoder, decoder, and conditional prior. It utilizes a Transformer-based text encoder combined with flow-based modules for spectrogram prediction, followed by HiFi-GAN vocoder-style transposed convolutional layers for final audio generation.
- Stochastic duration predictor for varied speech rhythms
- Flow-based module with coupling layers
- End-to-end training with variational lower bound and adversarial losses
- Non-deterministic output requiring seed fixing for reproducibility
Core Capabilities
- Direct text-to-speech synthesis for Yoruba language
- Variable speech rhythm generation
- High-quality spectrogram-based acoustic feature prediction
- Integrated with 🤗 Transformers library (v4.33+)
Frequently Asked Questions
Q: What makes this model unique?
This model is specifically trained for Yoruba language speech synthesis and uses a sophisticated VITS architecture that allows for natural variation in speech patterns through its stochastic duration predictor.
Q: What are the recommended use cases?
The model is ideal for applications requiring Yoruba language text-to-speech conversion, such as accessibility tools, educational software, or automated voice systems. It's particularly useful when natural-sounding speech variation is desired.
