myshell-ai/melotts

★ 7,418PythonAudience · developerComplexity · 2/5LicenseSetup · moderate

Mindmap

mindmap
  root((MeloTTS))
    What it does
      Text to speech
      Multiple languages
      Regional accents
    Languages
      English with accents
      Chinese-English mix
      Spanish French Korean Japanese
    Tech stack
      Python
      PyTorch
      HuggingFace models
    Use cases
      App voice output
      Accessibility
      Custom voice training

mindmap root((MeloTTS)) What it does Text to speech Multiple languages Regional accents Languages English with accents Chinese-English mix Spanish French Korean Japanese Tech stack Python PyTorch HuggingFace models Use cases App voice output Accessibility Custom voice training

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Add natural-sounding voice output to a Python app in English, Spanish, French, Chinese, Japanese, or Korean without needing GPU hardware.

USE CASE 2

Generate audio files from written text for use in videos, podcasts, e-learning content, or accessibility features.

USE CASE 3

Build a voice assistant or chatbot that speaks in a specific regional accent, British, Indian, Australian, or American English.

USE CASE 4

Create a multilingual text-to-speech pipeline that handles sentences mixing Chinese and English words in a single utterance.

Tech stack

PythonPyTorchHuggingFaceVITS

Getting it running

Difficulty · moderate Time to first run · 30min

Requires downloading pre-trained model files from HuggingFace, runs on CPU but a Python environment with PyTorch is needed.

MIT license, use freely for any purpose including commercial products, as long as you keep the copyright notice.

In plain English

MeloTTS is a text-to-speech library that converts written text into spoken audio. It was built by researchers at MIT and the company MyShell.ai, and it supports multiple languages and regional accents. Supported languages include English (with American, British, Indian, and Australian accent options), Spanish, French, Chinese, Japanese, and Korean. The Chinese model has a special feature: it can handle sentences that mix Chinese and English words in the same utterance. The library is designed to run fast enough for real-time use on a standard CPU, meaning you do not need expensive graphics hardware to generate speech with it. This makes it practical for developers building applications on ordinary machines or cloud servers without GPU resources. Users have three main ways to get started: trying it without any installation via a hosted option, installing it locally and using it through a Python API or command line, or training the system on a custom dataset to produce a different voice style. Pre-trained model files are hosted on HuggingFace, a common platform for sharing AI models. A web-based interface is also available for testing speech output interactively. The library is published under the MIT license, which means it is free to use in both personal projects and commercial products. The voice synthesis technology is built on top of earlier research systems called VITS and VITS2. The project was created by a small team from Tsinghua University and MIT, with community contributions adding the web and command-line interfaces. The README is short and focused on getting users started quickly rather than explaining the technical details of how the models work.

Copy-paste prompts

Prompt 1

I want to use MeloTTS to convert a paragraph of English text into speech with a British accent and save it as an MP3 file. Show me the Python code to do that from start to finish.

Prompt 2

Help me add MeloTTS to a FastAPI endpoint that accepts a JSON body with a text field and returns an audio file response the caller can play directly.

Prompt 3

I want to generate speech from a sentence that mixes Chinese and English words using MeloTTS. Show me how to load the Chinese model and pass a bilingual input string.

Prompt 4

Walk me through fine-tuning MeloTTS on a custom voice recording dataset to produce a different voice style, including what data format it expects.

Open on GitHub → Explain another repo

← myshell-ai on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.