Real-time AI noise suppression, transcription, and meeting recording

Privacy-first · Lightning-fast · Completely offline

Features • Quick Start • How It Works • Development

---

What is Crispy?

Crispy is a powerful desktop application that enhances your audio in real-time with AI-powered noise suppression, automatic transcription, and intelligent meeting recording—all without sending your data to the cloud.

Built with Tauri v2 (Rust backend + React/TypeScript frontend), Crispy processes everything locally on your machine, ensuring your conversations remain private while delivering professional-quality audio.

✨ Features

🎙️ Noise Suppression

• RNN Noise (nnnoiseless): real-time neural denoiser, 48 kHz, built-in model
• None / Test noise: passthrough or debug mode
• Adaptive volume control to maintain consistent levels
• Zero-latency monitoring for immediate feedback
• Routes clean audio to any application (Zoom, Discord, Teams, etc.)

📝 Smart Transcription

• Automatic transcription of all your recordings
• Chat with AI about your transcriptions using OpenAI/Poe.com
• Model metadata tracking—always know which model was used
• Chat history saved locally per recording

🎬 Meeting Recording

• Dual-source capture: Microphone + application audio (e.g., browser, Zoom)
• Stereo output: Mixed dual-mono for perfect playback
• Auto-resampling: Handles different sample rates (44.1kHz → 48kHz)
• One-click recording from system tray

🔒 Privacy-First Design

• 100% offline processing - no internet required for noise suppression
• Local storage - recordings, transcriptions, and settings stay on your machine
• Optional cloud AI - only for chat features, with your explicit API keys

🎨 Beautiful UX

• System tray popup for quick controls
• Dark mode ready with a clean, modern interface
• Settings persistence across dev/release builds

🚀 Quick Start

Prerequisites

• macOS 13.0+ (Ventura or later) for ScreenCaptureKit support
• Node.js 18+ and npm
• Rust (latest stable)
• Xcode Command Line Tools: xcode-select --install

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/crispy.git
   cd crispy

2. Install dependencies:

   make install

3. Install BlackHole 2ch (recommended):

   BlackHole is a virtual audio driver for routing processed audio to other apps.

   # Download from:
   https://existential.audio/blackhole/

4. Run in development:

   make dev

5. Build for production:

   make build

   The app bundle will be in target/aarch64-apple-darwin/release/bundle/

   Note: If you get invalid value '1' for '--ci' error, unset the CI variable:

   unset CI && make build

First-Time Setup

1. Grant microphone permissions when prompted
2. Grant screen recording permissions (required for app audio capture)
3. Select your microphone in Settings → General
4. Select BlackHole 2ch as output device (or any virtual audio device)
5. Choose a noise suppression model in the footer (General screen): None, Test noise, or RNN Noise (48 kHz mic recommended)
6. Optionally configure LLM in Settings → LLM Chat (for transcription chat)

🏗️ How It Works

Audio Pipeline

┌─────────────┐    ┌──────────────┐    ┌─────────────┐    ┌──────────────┐
│  Microphone │ -> │ Noise Model  │ -> │  Monitor    │ -> │  BlackHole   │ -> Apps
│  (Physical) │    │ RNNoise/None │    │  + Record   │    │  (Virtual)   │    (Zoom, etc.)
└─────────────┘    └──────────────┘    └─────────────┘    └──────────────┘
                                              │
                                              v
                                        ┌─────────────┐
                                        │   WAV File  │
                                        │  + App Audio│
                                        └─────────────┘

Technology Stack

Frontend

• React 18 + TypeScript
• Tailwind CSS v4 (modern design system)
• Tauri API for native integration
• Real-time event streaming

Backend (Rust)

• Audio: CPAL (cross-platform audio), Hound (WAV I/O), Rubato (resampling)
• Noise Suppression: nnnoiseless (RNNoise port, in ./rnnnoise), frame-based 48 kHz
• Transcription: transcribe-rs + ONNX Runtime (Whisper, Parakeet, Moonshine)
• App Audio Capture: ScreenCaptureKit (macOS 12.3+)
• LLM Integration: async-openai SDK for chat streaming
• Persistence: JSON-based settings + metadata

macOS Integration

• System tray with webview popup
• Window positioning via tauri-plugin-positioner
• Native permissions (Microphone, Screen Recording)

📦 Installation for Users

See detailed installation instructions: INSTALL.md

Quick start:

1. Download Crispy.app and move to /Applications/
2. Right-click on the icon → Open (not double-click!)
3. Confirm opening in the dialog

This bypasses macOS Gatekeeper for unsigned apps. After the first launch, you can open normally.

---

📖 Usage Guide

Basic Workflow

1. Start monitoring: Your mic is processed in real-time and routed to the output device
2. Start recording: Click record in settings or use the system tray
3. Stop recording: Audio is saved to ~/Documents/Crispy/Recordings/
4. View transcription: Auto-generated for each recording (if model selected)
5. Chat with AI: Ask questions about your transcription using LLM

Recording with App Audio

To record both your mic and application audio (e.g., Zoom calls):

1. Go to Settings → Recording → App Audio Capture
2. Select the application (Chrome, Zoom, etc.)
3. Start recording—both sources will be mixed in stereo

Transcription

1. Download a model: Settings → Transcription → Download models
2. Activate a model: Select it from the dropdown
3. Record: Transcription happens automatically after recording stops
4. View results: Click any recording to see the transcription window

LLM Chat (Optional)

Configure in Settings → LLM Chat:

• Endpoint: https://api.openai.com/v1 (or https://api.poe.com/v1)
• API Key: Your OpenAI/Poe API key
• Model: gpt-4, claude-3-5-sonnet, etc.

Once configured, chat with AI about your transcriptions to summarize, translate, or extract insights.

🛠️ Development

Project Structure

crispy/
├── src/                    # React/TypeScript frontend
│   ├── components/        # UI components
│   ├── hooks/             # React hooks (settings, models)
│   └── main.tsx           # Entry point
├── src-tauri/             # Rust backend
│   ├── src/
│   │   ├── main.rs        # Tauri app setup, audio pipeline, NS/TS models
│   │   ├── recording.rs   # Recording logic, ScreenCaptureKit
│   │   ├── commands/      # Tauri commands (models, ns_models, transcription, …)
│   │   └── managers/      # Transcription + model managers
│   ├── build.rs           # Build script (Swift linking)
│   └── Cargo.toml         # Rust dependencies
└── Makefile               # Build targets and commands

Useful Commands

make help                 # Show all available commands
make dev                  # Run in development mode
make build                # Build release binary
make clean                # Clean build artifacts
make install              # Install dependencies

Debugging

Enable Rust logs:

RUST_LOG=debug make dev

Check audio devices:

// In Rust backend
println!("Available devices: {:?}", devices);

🤝 Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

📄 License

MIT License - see LICENSE for details

---

_{Built with ❤️ using Rust, React, and Tauri}