ComfyUI Extension: ComfyUI-Soprano-TTS

Authored by SanDiegoDude

Created

Updated

2 stars

Ultra-lightweight, high-fidelity text-to-speech for ComfyUI using Soprano TTS.

Custom Nodes (0)

    README

    ComfyUI-Soprano-TTS

    Ultra-lightweight, high-fidelity text-to-speech for ComfyUI using Soprano TTS.

    Soprano TTS License PyTorch

    <img width="756" height="843" alt="image" src="https://github.com/user-attachments/assets/92aa3758-cef7-4977-86d7-209e12244bcf" />

    soprano_tts_00007.wav

    Features

    • Ultra-Fast: ~100-200x real-time generation
    • High-Fidelity: 32kHz crystal-clear audio
    • Lightweight: <1GB VRAM usage, 80M parameters
    • All-in-One: Integrated saving and playback with autoplay
    • ComfyUI Integration: Full memory management & interrupt support
    • Easy Installation: Works with PyTorch 2.5.1 (no ComfyUI breakage!)

    Quick Start

    Installation

    # 1. Navigate to ComfyUI and activate your Python environment
cd /path/to/ComfyUI
source venv/bin/activate  # Linux/Mac
# venv\Scripts\activate   # Windows

# 2. Install soprano-tts WITHOUT dependencies (critical!)
pip install soprano-tts --no-deps

# 3. Install required dependency
pip install unidecode

# 4. Restart ComfyUI

    That's it! The node uses transformers backend (scipy and transformers already in ComfyUI).

    Usage

    1. Add "Soprano TTS" node (audio/generation category)
    2. Enter text to synthesize
    3. Run workflow — audio plays automatically!
    4. (Optional) Connect audio output to other nodes for further processing

    First run: Downloads model (~500MB) from HuggingFace, takes ~30 seconds, then cached.

    Node Parameters

    Generation Settings

    | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | text | STRING | - | Text to convert to speech (2-15 sec sentences work best) | | temperature | FLOAT | 0.3 | Lower = consistent, higher = varied (0.0-2.0) | | top_p | FLOAT | 0.95 | Nucleus sampling threshold (0.0-1.0) | | repetition_penalty | FLOAT | 1.2 | Reduces repetition (1.0-2.0) | | seed | INT | -1 | Change to force regeneration (ComfyUI caching) | | keep_model_loaded | BOOLEAN | False | Keep in VRAM for faster subsequent runs |

    Audio Output Settings

    | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | filename_prefix | STRING | audio/soprano_tts | Path/prefix for saved files | | format | ENUM | wav | Output format: wav, mp3, flac, ogg | | bitrate | ENUM | 320k | Bitrate for mp3/ogg: 128k, 192k, 256k, 320k, VBR | | save | BOOLEAN | True | Save to output folder (False = temp/preview only) | | autoplay | BOOLEAN | True | Auto-play audio in the UI when generated |

    Output

    • audio (AUDIO): 32kHz waveform compatible with ComfyUI audio nodes
    • Built-in player: Audio plays directly in the node (with autoplay option)
    • Auto-save: Files saved to ComfyUI/output/audio/ with counters

    Tips for Best Results

    ✅ DO:

    • Use sentences 2-15 seconds long
    • Convert numbers to words ("1+1" → "one plus one")
    • Use proper grammar and contractions
    • Regenerate (change seed) if result isn't perfect

    ❌ DON'T:

    • Use very long or very short sentences
    • Include special characters (%, #, @, etc.)
    • Use all caps or improper grammar
    • Expect identical output each time (uses sampling)

    Why --no-deps Installation?

    The Problem: Soprano TTS officially requires PyTorch 2.8+, but ComfyUI uses PyTorch 2.5.1. Installing normally upgrades PyTorch and breaks ComfyUI with errors like:

    ImportError: undefined symbol: ncclMemFree

    The Solution: Install with --no-deps to skip the problematic lmdeploy dependency. Soprano automatically falls back to the transformers backend, which works perfectly with PyTorch 2.5.1!

    | Backend | Speed | PyTorch | Status | |---------|-------|---------|--------| | lmdeploy | ~2000x real-time | 2.8.0+ | ❌ Breaks ComfyUI | | transformers | ~100-200x real-time | 2.5.1+ | ✅ Works! |

    Even at 100-200x, generating 10 seconds of audio takes less than 1 second!

    Troubleshooting

    "No module named 'soprano'"

    pip install soprano-tts --no-deps
pip install unidecode

    ComfyUI won't start after installing

    Your PyTorch got upgraded accidentally. Fix it:

    # Remove conflicting packages
pip uninstall -y lmdeploy torch triton

# Reinstall correct PyTorch for ComfyUI
pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 \
  --extra-index-url https://download.pytorch.org/whl/cu121
pip install xformers==0.0.29.post1

    Model download fails

    • Check internet connection and HuggingFace access
    • Need ~500MB free disk space
    • Ensure huggingface-hub is installed: pip install huggingface-hub

    Poor audio quality

    • Lower temperature (try 0.2)
    • Convert numbers/symbols to words
    • Ensure sentences are 2-15 seconds
    • Regenerate with different seed

    Node not appearing in ComfyUI

    • Restart ComfyUI completely
    • Check console for loading errors
    • Run test: python test_compat.py

    Test Your Installation

    cd /path/to/ComfyUI/custom_nodes/ComfyUI-Soprano-TTS
python test_compat.py

    Should output: ✅ ALL TESTS PASSED! Soprano TTS should work!

    If tests fail, the script provides specific fix commands.

    Technical Details

    Compatibility Layer

    The node includes soprano_compat.py which:

    • Checks PyTorch version (requires 2.5.0+)
    • Patches torch.compiler.disable if needed (usually not needed for 2.5.1)
    • Auto-applies on import

    ComfyUI Integration

    • Memory Management: Properly loads/unloads models, cleans CUDA cache
    • Interrupt System: Respects user cancellation during generation
    • Audio Format: Returns standard ComfyUI AUDIO (waveform + sample_rate)
    • Model Caching: Downloads once, reuses forever

    Performance

    • Backend: transformers (HuggingFace)
    • Speed: ~100-200x real-time
    • Quality: 32kHz, perceptually indistinguishable from 44.1/48kHz
    • VRAM: <1GB (80M parameters)
    • Latency: First token in ~100ms, total generation <1s for 10s audio

    Limitations

    • No streaming mode (transformers backend limitation)
    • No voice cloning (Soprano limitation)
    • No multilingual support yet (Soprano limitation)
    • May occasionally mispronounce numbers/special characters

    Example Workflow

    See workflows/example_workflow.json for a complete example.

    Credits

    This node wraps the excellent Soprano TTS project:

    Soprano Acknowledgements

    Soprano uses and/or is inspired by:

    • Vocos - Vocoder architecture
    • XTTS - TTS inspiration
    • LMDeploy - Acceleration (optional)

    Requirements

    • OS: Linux or Windows
    • GPU: CUDA-enabled (CPU support coming to Soprano)
    • PyTorch: 2.5.0+ (tested with 2.5.1+cu121)
    • VRAM: <1GB
    • Disk: ~500MB for model cache
    • pydub: For integrated audio saving/playback (pip install pydub)
    • ffmpeg: Required by pydub for MP3/FLAC/OGG export

    License

    This ComfyUI node wrapper is licensed under Apache-2.0, matching the original Soprano project.

    Contributing

    Contributions welcome! Please:

    1. Fork the repository
    2. Create a feature branch
    3. Submit a pull request

    Support

    Changelog

    v1.0.0 (Initial Release)

    • Full Soprano TTS integration
    • PyTorch 2.5.1 compatibility via transformers backend
    • ComfyUI memory management & interrupt support
    • Automatic model downloading
    • Comprehensive error handling and testing

    Built with ❤️ for the ComfyUI community

    Ultra-fast, high-fidelity speech synthesis in your workflows!