🎙️ Maya1 TTS ComfyUI Custom Node

中文文档 | English

A high-quality text-to-speech (TTS) ComfyUI custom node powered by the Maya1 model, featuring multiple voice styles and flexible parameter configuration.

✨ Features

• 🎵 High-Quality Speech Synthesis: Based on the advanced Maya1 voice generation model
• 🎨 Multiple Voice Styles: 5 built-in voice presets (male, female, different ages)
• 🔧 Flexible Parameters: Adjustable temperature, chunk length, and more
• 💾 Automatic Model Management: Models auto-download and cache to ComfyUI models directory
• ⚡ Smart Caching: Models load once for improved generation efficiency
• 🌐 International Support: English UI and logging

📦 Installation

Method 1: Clone directly to ComfyUI custom nodes directory

cd ComfyUI/custom_nodes/
git clone https://github.com/ruanjianlun/comfyui_maya1_tts_alun.git
cd comfyui_alun_maya1
pip install -r requirements.txt

Method 2: Manual Installation

1. Copy the entire project folder to ComfyUI/custom_nodes/ directory
2. Install dependencies:

pip install -r requirements.txt

Dependencies

Main dependencies include:

• transformers - Hugging Face model library
• torch - PyTorch deep learning framework
• snac - SNAC audio codec
• soundfile - Audio file processing
• numpy - Numerical computing

🚀 Usage

1. Add Node in ComfyUI

1. Launch ComfyUI
2. Right-click on canvas, select Add Node -> audio -> maya1 -> Maya1 Text to Speech
3. The node will appear on the canvas

2. Configure Node Parameters

The node provides the following parameters:

3. Voice Style Presets

5 built-in voice style presets:

• Male-Mature: Male voice around 30 years old, American accent, normal pitch, warm timbre
• Female-Gentle: Female voice around 20 years old, American accent, slightly high pitch, soft timbre
• Male-Energetic: Male voice around 20 years old, American accent, high energy, bright timbre
• Female-Professional: Female voice around 30 years old, British accent, professional tone, clear timbre
• Neutral-Broadcast: Neutral voice, standard accent, clear articulation, balanced timbre

4. Custom Voice Description

If presets don't meet your needs, use the custom_description parameter for custom voice characteristics:

Realistic male voice in the 40s age with british accent. 
Deep pitch, authoritative tone, slow pacing.

Description template:

• Age: in the 20s/30s/40s age
• Accent: american/british/australian accent
• Pitch: high/normal/low/deep pitch
• Timbre: warm/bright/soft/clear timbre
• Pacing: fast/moderate/slow/conversational pacing
• Style: energetic/professional/gentle/authoritative tone

📁 Project Structure

comfyui_alun_maya1/
├── __init__.py              # Node registration entry file
├── maya_tts_node.py         # Main node implementation
├── config.py                # Configuration file (model paths, constants, etc.)
├── requirements.txt         # Python dependencies list
├── workflow_example.json    # ComfyUI workflow example
├── README.md               # Project documentation (English)
└── README.zh-CN.md         # Project documentation (Chinese)

🔧 Configuration

Model Storage Location

Models will auto-download to:

ComfyUI/models/maya1_tts_alun/

Includes the following model files:

• Maya1 main model (~4GB)
• Maya1 tokenizer
• SNAC audio decoder (~200MB)

Output File Location

Generated audio files are saved to:

ComfyUI/output/maya_tts_XXXXXX.wav

File naming format: maya_tts_{timestamp}.wav

Audio Specifications

• Sample Rate: 24000 Hz
• Format: WAV (16-bit PCM)
• Channels: Mono

📊 Workflow Example

The project includes a basic workflow example workflow_example.json, demonstrating how to:

1. Use the Maya1 TTS node to generate speech
2. Save the generated audio file
3. Preview and play the audio

Import Workflow

1. Open ComfyUI
2. Click the Load button
3. Select the workflow_example.json file
4. Click Queue Prompt to start generation

⚙️ Advanced Configuration

Modify Default Model Path

Edit the get_maya_model_path() function in config.py:

def get_maya_model_path():
    # Custom model path
    return "/your/custom/path/to/models"

Add New Voice Presets

Add to the VOICE_PRESETS dictionary in config.py:

VOICE_PRESETS = {
    # ... existing presets
    "Custom-Preset-Name": "Your custom voice description here",
}

Clear Model Cache

To free up VRAM, call:

from config import clear_model_cache
clear_model_cache()

🐛 Common Issues

Q1: First run is very slow?

A: First run requires downloading models (~4GB), please be patient. Models cache locally, subsequent usage will be much faster.

Q2: Out of memory errors?

A:

• Reduce the chunk_length parameter (e.g., change to 30)
• Close other VRAM-consuming programs
• Use CPU mode (auto-detects, no configuration needed)

Q3: Generated speech quality is not ideal?

A:

• Adjust the temperature parameter (0.3-0.5 usually works well)
• Try different voice presets
• Use custom descriptions for more precise voice characteristics
• Ensure input text has correct grammar and punctuation

Q4: Does it support Chinese?

A: Current version primarily supports English text-to-speech, Chinese support is limited. Recommended to use English text for best results.

Q5: How to speed up generation?

A:

• Ensure using GPU (CUDA)
• Reduce chunk_length (but may affect speech continuity)
• Lower temperature value
• Process long text in segments

🔄 Changelog

v1.0.0 (2025-01-09)

• ✨ Initial release
• 🎵 Support for Maya1 text-to-speech
• 🎨 5 built-in voice presets
• 🔧 Flexible parameter configuration
• 📁 Automatic model management
• 🌐 Complete English support

📄 License

This project is open source under the MIT License.

🙏 Acknowledgments

• Maya Research - Maya1 model
• Hugging Face - Model hosting
• ComfyUI - Workflow platform

📮 Contact

• Author: Alun
• Issues: Please submit on GitHub Issues

🌟 Support the Project

If this project helps you, please:

• ⭐ Star the project
• 🐛 Report bugs
• 💡 Suggest new features
• 🤝 Contribute code

---

Enjoy high-quality speech synthesis! 🎉