🎙️ Audio Reference - Generate speech using reference audio for voice characteristics
🗣️ Text-to-Speech - Generate natural-sounding speech from text
🎯 ASR Recognition - Automatic speech recognition for transcription
🔄 Batch Processing - Process multiple texts at once
🎨 Easy to Use - Simple ComfyUI nodes with clear inputs/outputs

Installation

Option 1: ComfyUI Registry (Recommended)

Install directly from ComfyUI Manager:

Open ComfyUI Manager
Search for "DN VoxCPM" or "VoxCPM"
Click Install
Restart ComfyUI

Option 2: Manual Installation

Clone this repository into your ComfyUI custom_nodes folder:

cd ComfyUI/custom_nodes/
git clone https://github.com/DekaNear/ComfyUI-DN-VoxCPM.git
cd ComfyUI-DN-VoxCPM
pip install -r requirements.txt

Restart ComfyUI after installation.

Available Nodes

Basic Nodes

1. VoxCPM Model Loader

Load the VoxCPM model for text-to-speech synthesis.

Inputs:

model_source: "huggingface" or "local"
model_id: Model ID (default: "openbmb/VoxCPM1.5")
enable_denoiser: Enable audio denoising
optimize: Use torch.compile for faster inference

Outputs:

model: Loaded VoxCPM model

Note: Models are downloaded automatically from HuggingFace on first use. The download may take a few minutes depending on your internet connection.

2. VoxCPM Text-to-Speech

Generate speech from text without audio reference.

Inputs:

model: VoxCPM model
text: Text to synthesize
cfg_value: Classifier-free guidance (1.0-3.0, default: 2.0)
inference_timesteps: Number of diffusion steps (4-30, default: 10)
normalize: Normalize output audio
retry_badcase: Retry if generation fails

Outputs:

audio: Generated audio

3. VoxCPM Audio Reference

Generate speech using a reference audio for voice characteristics.

Inputs:

model: VoxCPM model
text: Text to synthesize
prompt_audio: Reference audio (AUDIO type)
prompt_text: Transcription of reference audio
cfg_value: Classifier-free guidance (default: 2.0)
inference_timesteps: Diffusion steps (default: 10)
normalize: Normalize output
denoise_prompt: Denoise reference audio

Outputs:

audio: Generated audio with reference voice characteristics

4. VoxCPM Streaming TTS

Generate speech in streaming mode for real-time applications.

Inputs:

Same as Audio Reference, but optional prompt audio/text

Outputs:

audio: Streamed audio (concatenated)

5. VoxCPM Save Audio

Save generated audio to a WAV file.

Inputs:

audio: Audio to save
filename: Output filename (default: "output.wav")
output_dir: Optional custom output directory

Outputs:

None (saves to disk)

6. VoxCPM ASR Recognition

Automatic speech recognition for transcribing audio.

Inputs:

audio: Audio to transcribe (AUDIO type)

Outputs:

transcription: Recognized text

Advanced Nodes

7. VoxCPM Batch TTS

Process multiple texts at once with the same settings.

8. VoxCPM Batch Save Audio

Save multiple audio files from batch generation.

9. VoxCPM Concatenate Audio

Combine multiple audio clips with optional silence between them.

10. VoxCPM Text From File

Load text from .txt or .json files.

11. VoxCPM Model Info

Display information about the loaded model.

Utility Nodes

12. VoxCPM Load Audio

Load audio from file path and convert to ComfyUI AUDIO format.

Features:

Auto-searches in ComfyUI/input/ folder
Supports WAV, MP3, FLAC, OGG, and more
Automatic stereo to mono conversion
Just use filename if file is in ComfyUI/input/

13. VoxCPM Audio Converter

Convert audio from ComfyUI's LoadAudio to proper format for VoxCPM nodes.

Use case: Place between LoadAudio and VoxCPM nodes if you get "audio too short" errors.

Quick Start

Basic Text-to-Speech

[VoxCPM Model Loader] → [VoxCPM Text-to-Speech] → [VoxCPM Save Audio]

Using Audio Reference

Option 1: Using VoxCPM Load Audio (Recommended)

1. Put your audio file in ComfyUI/input/my_voice.mp3

2. Workflow:
[VoxCPM Model Loader]
        ↓
[VoxCPM Load Audio]
   audio_path: "my_voice.mp3"
        ↓
   ├─→ [VoxCPM ASR Recognition]
   │         ↓
   │   (transcription)
   │         ↓
   └─→ [VoxCPM Audio Reference] ← (model)
             ↓
      [VoxCPM Save Audio]

Option 2: Using ComfyUI LoadAudio

[LoadAudio] → [VoxCPM Audio Converter] → [VoxCPM ASR Recognition]
                                                ↓
                                         [VoxCPM Audio Reference]

Audio Loading Guide

Where to Put Audio Files

Recommended: Put your audio files in ComfyUI/input/ folder

Then in VoxCPM Load Audio, just write the filename:

my_audio.mp3

The node will find it automatically!

Alternative: Full Path

You can also use full paths:

Windows: C:/Users/YourName/audio/file.mp3
Linux/Mac: /home/user/audio/file.mp3

Supported Formats

✅ WAV (recommended)
✅ MP3
✅ FLAC
✅ OGG
✅ M4A
✅ Most common audio formats

Audio Requirements

For Audio Reference:

Duration: 3-10 seconds (ideal)
Quality: Clear voice, minimal background noise
Format: Any supported format

For ASR Recognition:

Minimum: 0.1 seconds
Recommended: 3-10 seconds
Content: Clear speech

Tips & Best Practices

Audio Quality: Better quality reference audio = better results
Audio Length: 3-10 seconds is ideal for audio reference
File Location: Put files in ComfyUI/input/ for easy access
Use ASR: Let ASR transcribe your reference audio automatically
Batch Processing: Use Batch TTS for multiple texts with same voice
Denoising: Enable denoiser for cleaner output

Configuration

Default settings are in config.py. You can modify:

Default CFG values
Inference timesteps
Model paths
Output directories

Requirements

Python 3.8+
PyTorch 2.0+
ComfyUI updated to latest version
See requirements.txt for full list

Credits

VoxCPM: OpenBMB/VoxCPM
ComfyUI: comfyanonymous/ComfyUI

License

Apache License 2.0

Support

For issues and questions:

Check this README
Check error messages in console
Open an issue on GitHub

Made by DekaNear
Version: 1.0.0
Last Updated: 2025-12-11