|
|
--- |
|
|
title: Mockingbird (v2 - current) |
|
|
app_file: app.py |
|
|
sdk: streamlit |
|
|
python_version: 3.11.8 |
|
|
streamlit_version: 1.44.1 |
|
|
pinned: false |
|
|
emoji: 🚀 |
|
|
colorFrom: green |
|
|
colorTo: green |
|
|
short_description: Generate speech from text in multiple languages. |
|
|
--- |
|
|
|
|
|
# Mockingbird TTS Demo |
|
|
This repo hosts Mockingbird, a demo of open Text-to-Speech tools. |
|
|
|
|
|
Currently, 3 synthesizers are supported: |
|
|
- [**Meta's Massively Multilingual Speech (MMS)**](https://ai.meta.com/blog/multilingual-model-speech-recognition/) model |
|
|
- [**Coqui's TTS**](https://docs.coqui.ai/en/latest/#) package and the models supplied via that |
|
|
- [**ESpeak-NG's**](espeak-ng) synthetic voices |
|
|
|
|
|
Voice conversion is achieved through Coqui. |
|
|
|
|
|
Notes: |
|
|
1. ESpeak-NG seems to have the worst performance out of the box, but it has a lot of options for controlling voice output. |
|
|
2. Coqui is no longer being officially developed. |
|
|
3. Where a synthesizer supports multiple models/voices, I manually pick the appropriate model. |
|
|
4. Not all synthesizers support a given language. |
