Pipeline
How it works
Each audio goes through 4 automatic stages, from input to the final formatted transcription.
Step 1
Audio Capture
Audio is captured via file upload, YouTube download (yt-dlp), or browser microphone via WebSocket.
Step 2
Speech-to-Text (Deepgram)
Audio is sent to the Deepgram Nova-3 API, which converts speech to text with high accuracy, supporting multiple languages.
Step 3
Raw Transcription
The raw text returned by Deepgram is saved. It includes timestamps, duration, and metadata from the original source.
Step 4
AI Processing (Groq)
DeepSeek via OpenRouter formats, organizes, and cleans the transcription. You can customize the processing prompt.
Speech-to-Text
Deepgram Nova-3
LLM
DeepSeek via OpenRouter
Realtime
WebSocket + Deepgram
Frontend
Next.js 16, React 19
Styling
Tailwind CSS 4