Parakeet Studio

Parakeet Studio is a local-first voice-to-text workstation built around the onnx-asr Parakeet v3 model paired with the latest Silero VAD ONNX detector. The experience combines a modern, glassmorphic interface inspired by the ElevenLabs UI system with a circular microphone visualizer for immersive real-time feedback.

Features

• 🎙️ Live microphone capture with RMS-based circular waveform visualizer and instant level feedback.
• 📁 File uploads for WAV, MP3, OGG, and WebM audio, automatically trimmed via Silero VAD before transcription.
• ⚙️ Rich settings to control VAD, punctuation restoration, speaker diarization heuristics, and language hints.
• 🪶 Parakeet v3 transcription pipeline with cached ONNX Runtime sessions and optional local storage of processed audio.
• 🧠 Silero VAD integration with configurable thresholds to detect speech segments accurately.
• 🗂️ Transcription timeline with copy-to-clipboard actions and metadata for each session.

Project layout

backend/   FastAPI application exposing transcription endpoints and ONNX-powered pipeline
frontend/  Vite + React (TypeScript) client with ElevenLabs UI-inspired styling

Getting started

Backend

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload

The first run downloads the Parakeet v3 and Silero VAD ONNX models into models/. Adjust locations via environment variables defined in app/config.py.

Frontend

cd frontend
npm install
npm run dev

The dev server defaults to http://localhost:5173 and expects the backend at http://localhost:8000 (configure via VITE_API_URL).

Notes

• The ONNX model URLs can be overridden by placing the model files at the paths defined in app/config.py before starting the server.
• To deploy behind HTTPS or enable GPU inference, update the FastAPI settings and the ONNX Runtime provider list in app/services/model_registry.py.