🎙️ Modal Transcriber MCP

中文版本

A powerful audio transcription streamhttp mcp server integrating Gradio UI, FastMCP Tools, and Modal cloud computing with intelligent speaker identification.

✨ Core Features

• 🎵 Multi-platform Audio Download: Support for Apple Podcasts, XiaoYuZhou, and other podcast platforms
• 🚀 High-performance Transcription: Based on OpenAI Whisper with multiple model support (turbo, large-v3, etc.)
• 🎤 Intelligent Speaker Identification: Using pyannote.audio for speaker separation and embedding clustering
• ⚡ Distributed Processing: Support for large file concurrent chunk processing, significantly improving processing speed
• 🔧 FastMCP Tools: Complete MCP (Model Context Protocol) tool integration
• ☁️ Modal Deployment: Support for both local and cloud deployment modes

🎯 Core Advantages

🧠 Intelligent Audio Segmentation

• Silence Detection Segmentation: Automatically identify silent segments in audio for intelligent chunking
• Fallback Mechanism: Long audio automatically degrades to time-based segmentation, ensuring processing efficiency
• Concurrent Processing: Multiple chunks processed simultaneously, dramatically improving transcription speed

🎤 Advanced Speaker Identification

• Embedding Clustering: Using deep learning embeddings for speaker consistency identification
• Cross-chunk Unification: Solving speaker label inconsistency issues in distributed processing
• Quality Filtering: Automatically filter low-quality segments to improve output accuracy

🔧 Developer Friendly

• MCP Protocol Support: Complete tool invocation interface
• REST API: Standardized API interface
• Gradio UI: Intuitive web interface
• Test Coverage: 29 unit tests and integration tests

🚀 Quick Start

Environment Setup

1. Clone Repository

git clone https://github.com/Sucran/modal-transcriber-mcp.git
cd modal-transcriber-mcp

2. Install Dependencies [Strongly Recommend Using uv]

uv init --bare --python 3.10
uv sync --python 3.12
source .venv/bin/activate

3. Configure Hugging Face Token (Optional, for speaker identification)

# Copy configuration template
cp config.env.example config.env
# YOUR_ACTUAL_TOKEN_HERE is your real Huggingface platform token
# This token needs permission to access the following three model repositories:
# pyannote/embedding: https://huggingface.co/pyannote/embedding 
# pyannote/segmentation-3.0: https://huggingface.co/pyannote/segmentation-3.0
# pyannote/speaker-diarization-3.1: https://huggingface.co/pyannote/speaker-diarization-3.1
sed -i 's/your-huggingface-token-here/YOUR_ACTUAL_TOKEN_HERE/' config.env

4. Modal Platform Authentication

# Need to login to Modal platform via web browser, then token will be saved locally
modal token new

5. Deploy Modal GPU Function Endpoints

python start_modal.py

Then modify your config.env:

MODAL_TRANSCRIBE_CHUNK_ENDPOINT=https://your-username--transcribe-audio-chunk-endpoint.modal.run
MODAL_HEALTH_CHECK_ENDPOINT=https://your-username--health-check-endpoint.modal.run
MODAL_GRADIO_UI_ENDPOINT=https://your-username--gradio-mcp-ui-app-entry.modal.run

Replace your-username with your actual Modal username

6. Local Deployment of Gradio and FastMCP (Optional, for local debugging/development)

python start_local.py

7. Modal Cloud Deployment of Gradio and FastMCP

modal deploy src.app::gradio_mcp_app

📚 How to Use This MCP Server

This application provides both Web Interface and MCP (Model Context Protocol) Tools for AI assistants to use.

Here's a demo video:

For local deployment, MCP configuration is:

{
    "mcpServers": {
        "podcast-mcp": {
            "url": "http://127.0.0.1:7860/api/mcp"
        }
    }
}

For Modal deployment, MCP configuration is:

{
    "mcpServers": {
        "podcast-mcp": {
            "url": "https://{your-username}--gradio-mcp-ui-app-entry.modal.run/api/mcp"
        }
    }
}

Both will use the GPU functions deployed on Modal: MODAL_TRANSCRIBE_CHUNK_ENDPOINT=https://{your-username}--transcribe-audio-chunk-endpoint.modal.run

🛠️ Technical Architecture

• Frontend: Gradio 5.31
• Backend: FastAPI + FastMCP
• Transcription Engine: OpenAI Whisper
• Speaker Identification: pyannote.audio
• Cloud Computing: Modal.com
• Audio Processing: FFmpeg

Future Plans

• Improve speaker identification accuracy
• Increase concurrent processing capacity of single GPU functions
• Optimize speaker clustering algorithms
• Support computing platforms in mainland China
• Test cost-effectiveness of other GPU types

🤝 Contributing

Issues and Pull Requests are welcome!

📜 License

MIT License

🔗 Related Links

• Test Coverage: 29 test cases ensuring functional stability
• Modal Deployment: Support for cloud high-performance processing