# Whisper Talk

> Push-to-talk voice transcription for macOS. Powered by OpenAI's Whisper AI, running 100% locally on your Mac. No cloud processing. No subscription. One-time purchase of $19.

## Key Facts

- **Platform**: macOS 12 (Monterey) and later — Apple Silicon (M1/M2/M3/M4) and Intel
- **Price**: $19 one-time purchase (no subscription, no recurring fees, no in-app purchases)
- **Privacy**: 100% offline — audio never leaves your Mac
- **AI Model**: OpenAI Whisper (tiny, base, small, medium models available)
- **Languages**: 50+ languages supported
- **Input Method**: Push-to-talk with customizable keyboard shortcut
- **Output**: Pastes transcribed text directly into any active application
- **AI Correction**: Optional local LLM post-processing for grammar and spelling
- **System Requirements**: 8GB RAM recommended (16GB for medium model with LLM correction)
- **Refund Policy**: 30-day money-back guarantee

## Features

### Push-to-Talk Voice Input
Hold a customizable keyboard shortcut to record your voice. Release to instantly transcribe and paste text into any active application. Works system-wide across all macOS apps — text editors, browsers, email clients, chat apps, IDEs, and more.

### 100% Local & Offline Processing
All speech recognition runs entirely on your Mac using OpenAI's Whisper model. After a one-time model download (75MB–1.5GB depending on model size), no internet connection is ever required for transcription. Your audio data never leaves your device.

### Multiple Whisper Model Sizes
Choose the model that fits your needs:
- **Tiny** (~75MB): Fastest, best for English
- **Base** (~150MB): Good balance of speed and accuracy
- **Small** (~500MB): Higher accuracy, multilingual
- **Medium** (~1.5GB): Near human-level accuracy for English

### AI-Powered Text Correction
Optional post-processing using local LLMs (Phi-3, Gemma) fixes grammar, punctuation, capitalization, and spelling while preserving your original meaning. Runs entirely on-device. Cloud providers (OpenAI, Anthropic) also available as an option.

### 50+ Language Support
Whisper supports over 50 languages including English, Spanish, French, German, Japanese, Chinese, Korean, Portuguese, Italian, Dutch, Russian, Arabic, Hindi, and many more. The base model and above are fully multilingual.

### Smart Clipboard Management
Whisper Talk preserves your existing clipboard contents and restores them after pasting transcribed text. Your workflow is never disrupted.

### Customizable Keyboard Shortcuts
Configure any keyboard shortcut for push-to-talk activation. Set up different shortcuts for different contexts or workflows.

### Menu Bar Integration
Lives quietly in your macOS menu bar. Minimal resource usage when idle. Instant activation when you need it.

### Universal App Compatibility
Works with any macOS application that accepts text input — Slack, VS Code, Safari, Chrome, Mail, Notes, Pages, Word, and every other app.

### Lightweight & Fast
Small app footprint. Instant startup. Transcription begins the moment you release the shortcut key. No loading screens or waiting.

## How It Works

1. **Press & Hold** your chosen keyboard shortcut to start recording audio from your microphone
2. **Speak naturally** — Whisper AI processes your speech locally on your Mac in real-time
3. **Release** — transcribed (and optionally corrected) text is instantly pasted into whatever application you're currently using

The entire process takes seconds. There is no need to copy-paste or switch between apps.

## Privacy Architecture

Whisper Talk is designed with privacy as a core principle:

- **No Cloud Processing**: All speech-to-text conversion happens on-device using the Whisper model
- **No Audio Storage**: Audio is processed in memory and immediately discarded after transcription
- **No Telemetry**: The app does not collect usage data, analytics, or crash reports
- **No Account Required**: No sign-up, no login, no email required to use the app
- **No Network Requests**: The app makes zero network requests during normal operation (unless you explicitly enable cloud LLM correction)
- **Local LLM Correction**: When AI correction is enabled with local models, text processing also stays entirely on-device
- **Clipboard Restoration**: Your clipboard is preserved and restored after each transcription paste
- **macOS Permissions**: Only requests Microphone and Accessibility permissions — nothing else

## Frequently Asked Questions

### What Mac do I need to run Whisper Talk?
Whisper Talk runs on macOS 12 (Monterey) and later, on both Apple Silicon (M1/M2/M3/M4) and Intel Macs. Apple Silicon Macs will see significantly faster transcription due to their neural engines. 8GB RAM is recommended for the base Whisper model; 16GB for the medium model with LLM correction.

### How accurate is the transcription?
Whisper is OpenAI's state-of-the-art speech recognition model. The base model handles most use cases excellently. The medium model (1.5GB) approaches human-level accuracy for English. With AI correction enabled, technical terms and proper nouns get cleaned up automatically.

### Does it really work 100% offline?
Yes. After you download Whisper Talk and your chosen model files (a one-time download of 75MB to 1.5GB depending on model size), everything runs locally. No internet connection is ever needed for transcription or correction.

### What languages does it support?
Whisper supports over 50 languages including English, Spanish, French, German, Japanese, Chinese, Korean, and many more. The base model and above are multilingual. The tiny model works best with English.

### How does it paste text into my apps?
Whisper Talk uses macOS accessibility APIs to paste text directly into your active application. On first launch, macOS will ask you to grant Accessibility permission. The app preserves your clipboard contents and restores them after pasting.

### What's the AI correction feature?
When enabled, a local LLM (like Phi-3 or Gemma) post-processes the transcription to fix grammar, punctuation, capitalization, and spelling while preserving your original meaning. It runs entirely on-device. You can also use cloud providers (OpenAI, Anthropic) if you prefer.

### Is there a subscription or recurring fee?
No. Whisper Talk is a one-time purchase of $19. You get the full app with all features, all model sizes, and lifetime updates. No subscription. No in-app purchases. No API keys required for local processing.

### Can I get a refund?
Yes, we offer a 30-day money-back guarantee. If Whisper Talk doesn't work for you, just reach out and we'll issue a full refund, no questions asked.

## Pricing Comparison

| Service | Price | Model | Internet Required | Privacy |
|---|---|---|---|---|
| **Whisper Talk** | $19 one-time | Local (Whisper) | No | Audio never leaves device |
| Otter.ai | $16.99/month ($203/year) | Cloud | Yes | Audio uploaded to cloud |
| Rev | $29.99/month ($359/year) | Cloud | Yes | Audio uploaded to cloud |
| Fireflies.ai | $19/month ($228/year) | Cloud | Yes | Audio uploaded to cloud |
| macOS Dictation | Free | Cloud/Local | Partial | Extended dictation uses cloud |

Whisper Talk pays for itself in the first month compared to any cloud-based subscription alternative, while keeping your data completely private.

## Links

- **Website**: https://whispertalk.app
- **Summary**: https://whispertalk.app/llms.txt