ComfyUI-Orpheus-TTS

This project adds high-quality Text-to-Speech capabilities to ComfyUI using the Orpheus TTS model. Create natural-sounding voices with emotional expressions, multilingual support, and audio effects.

Features

🎙️ High-quality, natural-sounding speech synthesis

🎭 Support for emotional expressions and paralinguistic elements

👥 Multiple voice options (tara, leah, jess, leo, dan, mia, zac, zoe, etc.)

📝 Long text handling with automatic chunking for consistent output

🎛️ Professional audio effects:

Pitch shifting (-12 to +12 semitones)

Speed adjustment (0.5x to 2.0x speed)

Volume control with anti-clipping protection

Audio normalization option

Reverb with adjustable room size and amount

Echo with configurable delay and decay

🌐 Optional support for private Hugging Face models

💻 Cross-platform: Works on Windows, Linux/WSL, and macOS

Installation

1. Install the Extension

Clone this repository into your ComfyUI’s custom_nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/ShmuelRonen/ComfyUI-Orpheus-TTS.git

2. Install Required Python Dependencies

pip install torch numpy soundfile transformers huggingface_hub nltk snac

For WSL 2, you may need to install directly from the GitHub repository:

pip install git+https://github.com/hubertsiuzdak/snac.git

3. Install SoX (Required for Audio Effects)

Windows

Download SoX for Windows from the official SourceForge page

Download the .exe installer (e.g., sox-14.4.2-win32.exe)

Run the installer:

Follow the installation prompts

Important: Note the installation directory (default is usually C:\Program Files (x86)\sox-14-4-2\)

No need to add to PATH – the extension uses the direct path to SoX

WSL 2 (Ubuntu)

sudo apt-get update
sudo apt-get install sox

macOS

brew install sox

4. Restart ComfyUI

After installing all required components, restart ComfyUI to load the extension.

Configuration

Hugging Face Authentication (Optional)

To access private models on Hugging Face, create a file named hf_config.json in the extension directory and insert your HF Token KEY:

“`json

{

“token”: “YOUR_HUGGING_FACE_TOKEN_HERE”

}

“`

Save the file and restart ComfyUI

Your token will be used to authenticate with Hugging Face when downloading models. This is only required if you’re using private models or if you need higher rate limits.

Nodes

Orpheus TTS Model Loader

Loads the required models for Orpheus TTS.

Inputs:

snac_model_path (optional): Path to SNAC model (default: “hubertsiuzdak/snac_24khz”)

orpheus_model_path (optional): Path to Orpheus model (default: “canopylabs/orpheus-3b-0.1-ft”)

Outputs:

model: Model reference to be passed to the generate node

Orpheus TTS Generate

Generates speech from text input.

Inputs:

model: Model reference from the loader node

text: The text to convert to speech

voice: Voice style to use (tara, leah, jess, leo, dan, mia, zac, zoe, etc.)

language (optional): Language for multilingual output (en, fr, es, etc.)

max_chunk_size (optional): Maximum chunk size for long text processing

Outputs:

audio: Audio data to be passed to preview or effects nodes

Orpheus Audio Effects

Applies high-quality audio processing to the generated speech.

Inputs:

audio: Audio data from the generate node

pitch_shift: Semitone adjustment (-12 to +12)

speed_factor: Playback speed modifier (0.5x to 2.0x)

sox_path (optional): Custom path to SoX executable

gain_db (optional): Volume adjustment in decibels

use_limiter (optional): Enable/disable limiter for positive gain

normalize_audio (optional): Enable/disable audio normalization

add_reverb (optional): Enable/disable reverb effect

reverb_amount (optional): Reverb intensity

reverb_room_scale (optional): Size of virtual space

add_echo (optional): Enable/disable echo effect

echo_delay (optional): Time between echo repetitions

echo_decay (optional): How quickly echo fades

Outputs:

audio: Processed audio data

Looking at the README section you provided, I’ll expand it to include information about the different element position options, including the new pipe feature:

Paralinguistic Elements

You can add expressive elements to the speech by inserting these tags:

– Natural laughter

– Light, subtle laughter

– Exhaling with emotion

– Clearing throat

– Subtle nasal sound

– Low, grumbling sound

– Tired exhale

– Sudden intake of breath

Element Position Options

The Element Position dropdown provides different ways to add these paralinguistic elements to your text:

None – No automatic element insertion. You can manually type the element tags in your text where desired.

“`

I can’t believe it! That’s amazing!

“`

Append – Automatically adds the selected element at the end of your text.

“`

Input: “That’s amazing!”

Output: “That’s amazing! “

“`

Prepend – Automatically adds the selected element at the beginning of your text.

“`

Input: “I need to get back to work.”

Output: “ I need to get back to work.”

“`

Pipe – Replace pipe characters (|) in your text with the selected element. This gives you precise control over element placement.

“`

Input: “I can’t believe it! | That’s the funniest thing | I’ve heard all day.”

Element: laugh

Output: “I can’t believe it! That’s the funniest thing I’ve heard all day.”

“`

Examples:

Manual placement (Element Position: None):

I can't believe it! <laugh> That's the funniest thing I've heard all day.
<sigh> But now I need to get back to work.

Using pipe placeholders (Element Position: Pipe):

Input: "Did you hear that? | It's hilarious! | I can't stop laughing!"
Element: laugh
Result: "Did you hear that? <laugh> It's hilarious! <laugh> I can't stop laughing!"

Multiple elements in one text:

<gasp> What was that? <pause> Did you hear something? <sigh> Maybe I'm just tired.

Audio Effect Tips

Volume Control

Gain Control: Use gain_db to increase or decrease volume without distortion

Positive values (0 to +20 dB): Increase volume with automatic clipping prevention

Negative values (-20 to 0 dB): Decrease volume

For best results with multiple effects, set gain last in your workflow

Normalization: Enable normalize_audio to automatically balance levels

Great for ensuring consistent volume across different voice samples

Applied before other effects for best results

Reverb

Reverb adds a sense of space to your audio. Here are some suggested settings:

Small Room: reverb_amount = 20, reverb_room_scale = 25

Medium Room: reverb_amount = 40, reverb_room_scale = 50

Large Hall: reverb_amount = 70, reverb_room_scale = 80

Cathedral: reverb_amount = 90, reverb_room_scale = 95

Echo

Echo creates repeating sound reflections. Good settings to try:

Subtle Echo: echo_delay = 0.3, echo_decay = 0.3

Moderate Echo: echo_delay = 0.5, echo_decay = 0.5

Canyon Echo: echo_delay = 1.0, echo_decay = 0.7

Effect Combinations

Phone Call: pitch_shift = 0, speed_factor = 1.0, add_reverb = True, reverb_amount = 10, reverb_room_scale = 10

Radio Announcer: pitch_shift = -2, speed_factor = 0.9, add_reverb = True, reverb_amount = 20, gain_db = 3

Stadium Announcement: pitch_shift = 0, speed_factor = 1.0, add_reverb = True, reverb_amount = 60, add_echo = True, echo_delay = 0.8

Child Voice: pitch_shift = 4, speed_factor = 1.1, gain_db = 2

Deep Voice: pitch_shift = -4, speed_factor = 0.9, gain_db = -2

Usage Examples

Basic Text-to-Speech

Add “Orpheus TTS Model Loader”

Add “Orpheus TTS Generate”

Connect the model loader’s output to the generate node’s input

Enter your text and select voice options

Connect to “Preview Audio” node to hear the result

Advanced: TTS with Audio Effects

Add “Orpheus TTS Model Loader”

Add “Orpheus TTS Generate”

Add “Orpheus Audio Effects”

Connect in sequence: Model Loader → Generate → Audio Effects → Preview Audio

Adjust pitch shift and speed factor sliders

Cross-Platform Compatibility

This extension has been tested and works on:

Windows 10/11

Linux (including WSL 2 on Windows)

macOS

Different environments may require specific setup steps:

Windows Notes

SoX is automatically located in standard installation directories

If installed elsewhere, provide the full path in the effects node

WSL 2 Notes

Use pip install git+https://github.com/hubertsiuzdak/snac.git to ensure compatibility

SoX is automatically located through the system PATH

macOS Notes

Install SoX via Homebrew for best compatibility

SoX Troubleshooting

Windows

If you encounter issues with SoX:

Verify the SoX path in the “Orpheus Audio Effects” node:

Default: C:\Program Files (x86)\sox-14-4-2\sox.exe

If your installation is in a different location, provide the full path to sox.exe

Check if SoX is installed correctly:

Open Command Prompt

Run "C:\Program Files (x86)\sox-14-4-2\sox.exe" --version

If you get an error, reinstall SoX

WSL 2 (Ubuntu)

Verify SoX installation:

“`bash

sox –version

“`

If SoX is not found, install it:

“`bash

sudo apt-get update

sudo apt-get install sox

“`

Model Details

This extension uses the following models:

Orpheus TTS: A powerful text-to-speech model developed by Canopy AI

SNAC Codec: A high-quality neural audio codec for voice synthesis

License

This project uses models with their own licenses:

Orpheus Model: Canopy AI’s Orpheus-TTS

SNAC Model: Hubert Siuzdak’s model

Please consult these licenses for usage terms and restrictions.

Credits

Original Orpheus TTS implementation by Canopy AI

SoX audio processing library: SoX – Sound eXchange

ComfyUI: ComfyUI