README_API.md

metadata

title: AI Voice Enhancer API
emoji: 🎙️
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 5.50.0
app_file: app_wrapper.py
pinned: false

🎙️ AI Voice Enhancer API

Professional podcast-quality audio enhancement powered by AI. This Space provides a REST API for audio processing.

🚀 Quick Start

For End Users (with Frontend)

1. Download the frontend from the repository
2. Open frontend/index.html in your browser
3. Configure API URL to this Space: https://YOUR-SPACE-NAME.hf.space
4. Upload audio and enhance!

For Developers (API Integration)

curl -X POST "https://YOUR-SPACE-NAME.hf.space/enhance" \
  -F "file=@input.wav" \
  -F "speech_level=80" \
  -F "background_level=85" \
  -o enhanced.wav

📡 API Endpoints

Endpoint	Method	Description
/	GET	API information
/health	GET	Health check
/enhance	POST	Enhance audio file
/cleanup	DELETE	Clean temp files

🎚️ Parameters

POST /enhance

• file: Audio file (WAV, MP3, M4A, FLAC, OGG, AAC)
• speech_level: 0-100 (default: 80) - Voice clarity
• background_level: 0-100 (default: 85) - Noise reduction

🎯 Features

• ✅ Noise Suppression: Facebook Denoiser (DNS) + FFmpeg
• ✅ Clarity Enhancement: Multi-stage EQ, compression, de-essing
• ✅ Loudness Normalization: EBU R128 standard (-16 LUFS)
• ✅ Adjustable Processing: Fine-tune enhancement levels
• ✅ CPU Optimized: Runs on free Spaces
• ✅ CORS Enabled: Works with any frontend

🔧 Technical Details

Audio Processing Pipeline:

1. Validation - Check format, duration, file size
2. Noise Suppression - DNS model or FFmpeg afftdn
3. Clarity Enhancement - EQ, gate, compression, de-esser
4. Normalization - EBU R128 loudness standard

Processing Time:

• 10-second audio: ~15-30 seconds
• 60-second audio: ~60-120 seconds
• Depends on CPU availability

📊 Quality

Targets 80-90% of Adobe Podcast Enhance Speech quality:

Feature	Adobe Podcast	This API
Noise Removal	⭐⭐⭐⭐⭐	⭐⭐⭐⭐
Clarity	⭐⭐⭐⭐⭐	⭐⭐⭐⭐
Speed	⭐⭐⭐	⭐⭐⭐⭐
Cost	💰💰💰	Free ✅

💻 Example: Python Client

import requests

class VoiceEnhancer:
    def __init__(self, api_url):
        self.api_url = api_url
    
    def enhance(self, audio_file, speech=80, background=85):
        url = f"{self.api_url}/enhance"
        files = {"file": open(audio_file, "rb")}
        data = {
            "speech_level": speech,
            "background_level": background
        }
        
        response = requests.post(url, files=files, data=data)
        
        if response.ok:
            return response.content
        else:
            raise Exception(f"Enhancement failed: {response.text}")

# Usage
enhancer = VoiceEnhancer("https://YOUR-SPACE-NAME.hf.space")
enhanced_audio = enhancer.enhance("input.wav", speech=80, background=85)

with open("output.wav", "wb") as f:
    f.write(enhanced_audio)

📝 Limitations

• Duration: Max 60 seconds
• File Size: Max 50MB
• Concurrent Users: Limited on free tier
• Processing Speed: Depends on CPU availability

🛠️ Tech Stack

• API Framework: FastAPI + Uvicorn
• Audio Processing:
  • PyTorch + Torchaudio
  • Facebook Denoiser (DNS)
  • FFmpeg (filters, normalization)
• ML Model: DNS64 (pretrained)

📦 Deployment

This Space runs:

• FastAPI server on port 7860
• Gradio info page on port 7861 (this page)

The actual audio processing happens via the FastAPI endpoints.

🔒 Privacy

• No data stored permanently
• Temporary files auto-deleted after 30 minutes
• No analytics or tracking
• Processes on-server (no external APIs)

📄 License

MIT License - Free to use, modify, and distribute.

🙏 Acknowledgments

• Facebook AI: For Denoiser library
• FFmpeg: For robust audio tools
• Hugging Face: For free hosting

---

Repository: [GitHub Link Here]
Frontend: Download from repository
Support: Open an issue on GitHub