Skip to content

README.md

Latest commit

 

History

History
173 lines (126 loc) · 6.36 KB

README.md

File metadata and controls

173 lines (126 loc) · 6.36 KB
Raw

SpeakIDE

Voice dictation for IntelliJ IDEA — press a shortcut, speak, and your words appear wherever the cursor is: code editor, AI Chat, Junie, commit message, any text field.

Demo

Demo Video

Features

  • Universal input — works in any text field: code editor, AI Chat / Junie, commit message, search boxes
  • Three STT backends — switch between cloud and offline in settings:
    • OpenAI Whisper (cloud) — high accuracy, supports any OpenAI-compatible endpoint (OpenAI, Groq, etc.)
    • Whisper Local (offline) — runs a ggml Whisper model on-device via whisper.cpp JNI, no internet required
    • Vosk (offline) — fully local, lightweight, no internet required, no data leaves your machine
  • Auto-stop on silence — recording ends automatically after a configurable pause; no need to press the shortcut again
  • Clipboard fallback — if no editor is focused when recording stops, the transcribed text is copied to clipboard
  • Microphone button in AI Chat toolbar — one-click dictation directly from the AI Assistant / Junie input bar
  • Configurable language — set a language hint or leave on auto for automatic detection
  • Secure key storage — API key stored in IntelliJ's PasswordSafe, never in plain text on disk

Installation

From JetBrains Marketplace (recommended)

  1. Open Settings → Plugins → Marketplace
  2. Search for SpeakIDE
  3. Click Install and restart the IDE

Or install directly: plugins.jetbrains.com/plugin/31433-speakide

From source

Requirements: JDK 21, IntelliJ IDEA 2024.2+

git clone https://github.com/Dragonav4/SpeakIDE.git
cd SpeakIDE
./gradlew runIde          # launch a sandbox IDE with the plugin
./gradlew buildPlugin     # build distributable .zip → build/distributions/

Install the .zip via Settings → Plugins → ⚙ → Install Plugin from Disk.

Setup

OpenAI Whisper (cloud)

  1. Open Settings → Tools → SpeakIDE
  2. Select Provider → OpenAI Whisper (Cloud)
  3. Paste your API key (stored securely)
  4. Set Base URL and Model Name:
Provider Base URL Model
OpenAI https://api.openai.com/v1 whisper-1
Groq https://api.groq.com/openai/v1 whisper-large-v3-turbo

Vosk (offline)

  1. Download a model from alphacephei.com/vosk/models
    Recommended: vosk-model-small-en-us (~50 MB) or vosk-model-small-ru
  2. Extract the archive
  3. Open Settings → Tools → SpeakIDE
  4. Select Provider → Vosk (Offline)
  5. Set Vosk model path to the extracted folder

Whisper Local (offline)

  1. Download a ggml model from Hugging Face
    Recommended: ggml-base.en.bin (~150 MB) or ggml-small.bin for multilingual
  2. Open Settings → Tools → SpeakIDE
  3. Select Provider → Whisper (Local)
  4. Set Whisper local model (.bin) to the downloaded .bin file

Note for macOS Apple Silicon users: the bundled native library supports both arm64 and x86_64 — no extra steps needed.

Minimum audio length: Whisper Local requires at least ~1 second of speech to produce a result. Very short recordings (under 1 s) will return empty output.

Usage

Action How
Start / stop recording Ctrl+Alt+/ (all platforms)
Start / stop from AI Chat Click the 🎤 button in the AI Chat input toolbar
Auto-stop Speak, then pause — recording stops after silence timeout
No editor focused Transcribed text is copied to clipboard automatically

The microphone icon in the toolbar turns red while recording is active.

Settings Reference

Settings → Tools → SpeakIDE

Setting Description
Provider OpenAI Whisper (Cloud), Vosk (Offline), or Whisper (Local)
Language Language hint (auto, en, ru, de, …) — Vosk ignores this
API Key Whisper cloud API key — stored in PasswordSafe
Base URL Whisper-compatible endpoint URL
Model Name Model identifier for the cloud API
Vosk model path Path to the extracted Vosk model folder
Whisper local model (.bin) Path to a ggml .bin model file
Enable silence detection Auto-stop recording after a pause
Silence threshold (ms) How long the pause must be before stopping (default 2000 ms)
Show recording overlay Visual indicator while recording

Architecture

ToggleRecordingAction  (~30 lines, delegates only)
        │
        ▼
RecordingService  (@Service APP — owns the state machine)
        │
        ├── RecordingState (sealed class: Idle / Recording / Transcribing)
        │
        ├── AudioSource (interface)
        │       └── AudioCapture (TarsosDSP)
        │               ├── MicPermissionProbe
        │               └── SilenceTimeoutProcessor
        │
        ├── PcmBuffer  (ConcurrentLinkedQueue — thread-safe PCM accumulator)
        │
        ├── SttProviderFactory
        │       ├── OpenAiWhisperProvider  ── shared HttpClient (lazy) ── Whisper API
        │       ├── WhisperLocalProvider   ── whisper.cpp JNI ── ggml model (offline)
        │       │       └── AudioResampler  (44 100 Hz → 16 kHz, TarsosDSP sinc)
        │       └── VoskProvider  ── Vosk JNI (native arm64 / x86_64)
        │
        ├── TextDelivery (fun interface)
        │       ├── CaretTextDelivery   (WriteCommandAction on EDT)
        │       └── ClipboardTextDelivery  (fallback when no editor focused)
        │
        ├── RecordingIndicator (interface)
        │       └── RecordingOverlay  (floating Swing window)
        │
        └── SpeakIdeNotifier  (wraps NotificationGroupManager)

The AI Chat microphone button is a second registration of ToggleRecordingAction added to the AI Assistant toolbar — loaded only when the com.intellij.ml.llm plugin is present (optional dependency).

Contributing

Pull requests are welcome. For major changes, open an issue first.

./gradlew test            # run unit tests
./gradlew runIde          # test in sandbox IDE

License

APACHE 2.0