Record yourself speaking and hear a cleaned-up version played back in your own cloned voice — filler words removed, phrasing tightened.

The demo moment: A volunteer speaks for 60 seconds. Within 30 seconds they hear themselves back, clean and confident.

1. Voice Sample — Record a 30-second reading to clone your voice
2. Main Recording — Record your actual speech (up to 5 minutes)
3. Configure — Pick target audience and style (optional)
4. Process — ElevenLabs Scribe transcribes → GPT-4o cleans → ElevenLabs speaks in your voice
5. Results — Side-by-side original vs. cleaned audio, transcript diff, filler stats

• Node.js 18+
• Python 3.12+
• OpenAI API key
• ElevenLabs API key

cd backend
cp .env.example .env
# Fill in OPENAI_API_KEY and ELEVENLABS_API_KEY in .env

python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

cd frontend
npm install
npm run dev

Open http://localhost:3000.

├── backend/
│   ├── main.py               # FastAPI app — /health, /clone, /analyze
│   ├── stt_client.py         # ElevenLabs Scribe transcription
│   ├── gpt_client.py         # GPT-4o filler removal
│   ├── elevenlabs_client.py  # Voice clone + TTS
│   └── requirements.txt
└── frontend/
    ├── app/
    │   ├── page.tsx           # 3-step recording + config flow
    │   └── results/page.tsx   # Stats, dual audio players, transcript diff
    └── components/
        ├── Recorder.tsx        # MediaRecorder with MIME detection
        ├── AudioPlayer.tsx     # Labeled audio player with download
        └── TranscriptDiff.tsx  # Filler word highlighting

Form fields: audio, voice_id, audience (General/Investors/Technical), style (Neutral/More Confident/Add Humor), duration

Response:

{
  "raw_transcript": "...",
  "cleaned_transcript": "...",
  "fillers": [{ "word": "um", "count": 12 }],
  "total_fillers": 20,
  "original_wpm": 145,
  "cleaned_wpm": 132,
  "audio_url": "/audio/abc123.mp3"
}