Ez

th1lo

Desktop app for local audio recording, Whisper transcription, and word-level annotation. Built with Tauri 2 + Svelte 5.

EZ

A desktop app for audio recording, transcription, and annotation. Record voice, capture screenshots and clipboard content simultaneously, get word-level transcripts with timestamps, and annotate them with rich metadata — all processed locally, no cloud APIs required.

Built with Tauri 2 (Rust backend) + Svelte 5 (TypeScript frontend).

Features

Recording & Capture

Select any audio input device with real-time amplitude visualization
Automatically capture screenshots from a watched folder during recording
Automatically capture clipboard images every 1.5 seconds during recording
All captures are timestamped and associated with the recording session

Transcription

Offline transcription via Whisper (large-v3-turbo) with Metal GPU acceleration on macOS
Word-level timestamps using Dynamic Time Warping (DTW)
Voice Activity Detection (VAD) for improved accuracy on silent segments
Model downloaded automatically on first use (~874 MB, stored locally)

Annotation

Annotate any word range in the transcript with:
- Footnotes
- Hyperlinks
- Code snippets (with language tag)
- Images
- Audio clips
Conflict detection prevents overlapping annotations
Soft delete with undo (3-second toast window)

Session Management

Sessions persist across app restarts; last opened session is restored automatically
Rename and delete sessions
Export any session as a structured directory: session.json, transcript.json, annotations.json, plus all asset files

Tech Stack

Layer	Technology
Desktop framework	Tauri 2
Frontend	Svelte 5 + SvelteKit 2
Language	TypeScript (frontend), Rust (backend)
Transcription	whisper-rs with Metal acceleration
Audio I/O	cpal + hound
Audio resampling	rubato
Database	SQLite via sqlx + tauri-plugin-sql
File watching	notify
Build tool	Vite 8

Prerequisites

Node.js 20+
Rust 1.77+
Tauri CLI prerequisites for your OS
- macOS: Xcode Command Line Tools
- Linux: libwebkit2gtk, libgtk-3, libappindicator3, etc.
- Windows: Microsoft Visual C++ Build Tools

Development

# Install frontend dependencies
npm install

# Start dev mode (launches Vite dev server + Tauri window)
npm run tauri dev

The first build will compile all Rust dependencies, which takes a few minutes. Subsequent builds are incremental.

Building

# Build frontend
npm run build

# Bundle desktop app (outputs to src-tauri/target/release/bundle/)
npm run tauri build

Output formats by platform:

macOS: .dmg and .app
Windows: .msi and .exe
Linux: .AppImage and .deb

Project Structure

ez/
├── src/                    # Svelte frontend
│   ├── lib/
│   │   ├── components/     # UI components (AudioPlayer, TranscriptView, etc.)
│   │   ├── db.ts           # SQLite schema and initialization
│   │   ├── audio.ts        # Recording logic + device enumeration
│   │   ├── transcript.ts   # Transcript data access
│   │   ├── annotations.ts  # Annotation CRUD
│   │   ├── sessions.ts     # Session management
│   │   ├── captures.ts     # Clipboard + watcher capture handling
│   │   └── export.ts       # Session export
│   └── routes/
│       ├── +page.svelte    # Home (session list + recording controls)
│       └── session/[id]/   # Session detail view (transcript + annotations)
├── src-tauri/
│   ├── src/
│   │   ├── lib.rs          # Tauri app setup, command registration, URI schemes
│   │   ├── audio.rs        # Audio recording, device management
│   │   ├── transcribe.rs   # Whisper inference
│   │   ├── model.rs        # Model download and validation
│   │   ├── sessions.rs     # Session backend commands
│   │   ├── annotations.rs  # Annotation backend commands
│   │   ├── watcher.rs      # File system watcher
│   │   ├── resample.rs     # Audio resampling
│   │   └── export.rs       # Export command
│   ├── capabilities/
│   │   └── default.json    # Tauri permission grants
│   ├── Cargo.toml
│   └── tauri.conf.json
├── package.json
└── svelte.config.js

Data Storage

All app data is stored locally in the platform-specific app data directory:

Platform	Path
macOS	`~/Library/Application Support/com.ez.app/`
Windows	`%APPDATA%\ez\`
Linux	`~/.config/ez/`

Contents:

ez.db — SQLite database (sessions, transcripts, words, annotations)
recordings/ — WAV audio files
captures/ — Clipboard/screenshot captures
models/ — Downloaded Whisper and VAD model weights

Database Schema

Table	Purpose
`sessions`	Recording sessions with metadata (duration, device, sample rate)
`transcripts`	One per session; stores aggregated word count and detected language
`words`	Individual words with `start_ms` / `end_ms` timestamps
`annotations`	Rich annotations tied to word index ranges; supports soft delete
`pending_captures`	Staging area for clipboard/watcher assets before user annotation
`settings`	Key-value store (watch folder path, last open session)

Security

Audio and asset files are served via custom URI schemes (audio://, asset://) with path canonicalization to prevent directory traversal
No data is sent to external services; transcription and all processing runs locally
Model downloads use HuggingFace CDN with minimum-size validation to guard against truncated files

macOS Permissions

On first launch, macOS will request:

Microphone — required for audio recording
Files and Folders — required for watch folder access and asset storage

These are granted via standard macOS permission dialogs and can be revoked in System Settings > Privacy & Security.

Top categories