Spaces:
Running on Zero
Running on Zero
File size: 1,722 Bytes
99ced08 6784f71 99ced08 6784f71 99ced08 6784f71 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 | ---
title: Maya 2 Native TTS
emoji: 🗣️
colorFrom: blue
colorTo: pink
sdk: gradio
sdk_version: 6.30.0
app_file: app.py
python_version: "3.10"
short_description: Streaming multilingual TTS with zero-shot voice cloning
startup_duration_timeout: 1h
---
# Maya 2 Native — multilingual streaming TTS
Demo of [maya-research/maya2-native](https://huggingface.co/maya-research/maya2-native):
a 2.5B-parameter text-to-speech model covering **12 Indic languages + English**
(Hindi, English, Assamese, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia,
Punjabi, Tamil, Telugu) with **zero-shot voice cloning** from a 3–10 s reference
clip. Apache-2.0. Output is 24 kHz mono audio.
- **Named voices**: 20 trained voices ship with the release — pick one from the
dropdown, or leave "(none)" for the plain (unclaimed) voice.
- **Voice cloning**: upload a 3–10 s clip plus its exact transcript; the same
speaker reads your text — including in a *different* language (cross-lingual
cloning).
- **Text normalisation** (on by default): expands digits, currency, dates, URLs
and ALL-CAPS names into spoken words in the target language.
This Space vendors the [maya3](https://github.com/MayaResearch/maya2-native)
inference engine (`maya3/`, `tts/`, `serve/`, `scripts/`, `configs/`) under its
Apache-2.0 license. Model weights (Mimi codec, ECAPA speaker encoder) are
downloaded from the Hugging Face cache on first use.
## Examples
The example rows are the showcase texts from the model card, one per supported
language, each paired with its native voice. The bundled
`samples/hi_ritvik.wav` is a reference clip from the model repository you can
use to try voice cloning (its transcript is in the model card).
|