EchoForge is an advanced voice generation platform that allows you to create lifelike vocal content with customizable parameters.
For troubleshooting API issues, a new debug page has been added. To access it:
- Start the server as described below
- Visit:
http://localhost:8765/debug - Use the debug interface to test different API endpoints and trace request/response flow
https://github.com/toddllm/echoforge
- Text-to-speech generation with multiple character voices
- Adjustable generation parameters (temperature, top-k, style)
- Background task processing for long-running generations
- RESTful API for integration with other applications
- Web interface with light and dark mode support
- Easy environment configuration
EchoForge is built on top of the Conversational Speech Model (CSM) from Sesame AI Labs. CSM is a speech generation model that generates high-quality, natural-sounding speech from text input. The model architecture employs a Llama backbone and a smaller audio decoder that produces Mimi audio codes.
Key features of the CSM model:
- High-quality speech synthesis
- Support for multiple speakers
- Contextual awareness for more natural-sounding conversations
- Adjustable generation parameters (temperature, top-k)
EchoForge wraps this technology in a user-friendly web interface and API, making it accessible for various applications.
EchoForge now uses the Direct CSM implementation by default for faster voice generation, especially on CUDA-enabled devices.
- Up to 25x faster generation on GPU compared to CPU
- Maintains the same high-quality voice output
- Automatically falls back to CPU if CUDA is unavailable
To start the server with Direct CSM enabled:
python run.py --direct-csm
When using the API, specify device=cuda for faster generation:
curl -X POST http://localhost:8779/api/generate \
-H "Content-Type: application/json" \
-d '{"text": "Your text here", "voice": "male", "temperature": 0.7, "device": "cuda"}'
- CUDA generation: ~3-6 seconds
- CPU generation: ~150 seconds
uv is a faster, more reliable Python package installer and resolver.
- Install uv:
pip install uv- Clone the repository:
git clone https://github.com/yourusername/echoforge.git
cd echoforge- Create a virtual environment and install dependencies:
uv venv
source .venv/bin/activate # On Windows: .venv\Scripts\activate
uv pip install -r requirements.txt- Clone the repository:
git clone https://github.com/yourusername/echoforge.git
cd echoforge- Create a virtual environment and install dependencies:
python -m venv .venv
source .venv/bin/activate # On Windows: .venv\Scripts\activate
pip install -r requirements.txt- Create a
.envfile in the project root with the following content:
DEBUG=true
HOST=0.0.0.0
PORT=8765
OUTPUT_DIR=./static/voices
STATIC_DIR=./static
TEMPLATES_DIR=./templates
- Configure voice directories:
# Create directories for voice data
mkdir -p static/voices/creativeUse the provided script to start the server:
./start_server.shOr manually:
uvicorn run:app --host 0.0.0.0 --port 8765 --reloadUse the provided script to stop the server:
./stop_server.shThe debug page is a powerful tool for troubleshooting voice generation issues:
- Navigate to
http://localhost:8765/debugin your browser - The debug page provides:
- API selection between
/api/v1/generateand/api/voices/generate - Detailed request/response information
- Task ID validation and tracking
- Audio URL construction debugging
- Network connection testing
- API selection between
- API Version Selection: Test both API versions to identify compatibility issues
- Parameter Controls: Adjust voice generation parameters like temperature and top-k
- Request Inspection: View the exact request payload being sent
- Response Monitoring: See the raw response and parsed data
- Status Checking: Monitor the task progress through different endpoints
- URL Validation: Troubleshoot issues with audio URL construction
- Network Testing: Test if endpoints are accessible and responding
-
POST /api/v1/generate: Generate voice using API v1{ "text": "Text to convert to speech", "speaker_id": 1, "temperature": 0.7, "top_k": 50, "device": "auto" } -
POST /api/voices/generate: Generate voice using voices API{ "text": "Text to convert to speech", "speaker_id": 1, "temperature": 0.7, "top_k": 50, "device": "auto" }
GET /api/v1/tasks/{task_id}: Check task status with API v1GET /api/voices/tasks/{task_id}: Check task status with voices API
GET /api/voices: Get list of available voicesGET /api/voices/list: Alternative endpoint for voice listing
-
"GET http://localhost:8765/undefined 404 (Not Found)"
- This indicates an issue with audio URL construction
- Use the debug page to trace the URL construction process
- Ensure task IDs are properly validated
-
Speaker ID validation errors
- If using
speaker_id: 0, the system will automatically map tospeaker_id: 1 - Verify speaker IDs in the range 1-5 are available
- If using
-
API endpoint not found
- The system supports both
/api/v1/generateand/api/voices/generate - Use the debug page to test which endpoint is working
- Check for t 7D32 ypos in endpoint URLs
- The system supports both
-
Voice files not playing
- Ensure the
static/voices/creativedirectory contains voice files - Check the audio URL construction in the debug logs
- Verify file permissions on the voice files
- Ensure the
The debug page contains a JavaScript console that provides detailed logging:
- Open your browser's developer tools (F12 or right-click -> Inspect)
- Navigate to the Console tab
- Look for errors or warnings related to API calls
- The debug page will also display detailed logs in its interface
Please read CONTRIBUTING.md for details on our code of conduct and the process for submitting pull requests.
This project is licensed under the MIT License - see the LICENSE file for details.
- This project uses the CSM model architecture from Sesame AI Labs for voice generation
- Special thanks to the open-source community for their contributions to speech synthesis technology
The EchoForge admin panel includes a simplified voice generation interface that allows you to:
- Enter text to be spoken
- Select a voice (Male, Female, or Child)
- Adjust temperature settings
- Generate and download audio files
To access the voice generation interface:
- Navigate to
http://[server-address]:[port]/admin/voices - Log in with your admin credentials
- Use the form to generate voice samples
The interface provides real-time feedback on generation status and device information during processing.