BetterWispr: Open-Source, Local Dictation for macOS

BetterWispr is a privacy-first, free tool designed for macOS that allows you to "speak messy" and receive polished text. It eliminates filler words, corrects false starts, and adapts its writing style based on the application you are using.

🚀 Quick Start & Compatibility

• OS Requirement: macOS 14+
• Hardware: Apple Silicon (Recommended) or Intel
• Cost: Free & Open Source (Apache 2.0)
• Privacy: No accounts, no cloud fallback, no data uploads.

---

🛠 Core Functionality

The Workflow

BetterWispr transforms your spoken words into clean text without ever leaving your machine.

⌨️ Control & Interaction

The default trigger is ⌥ + Space. You can choose between two modes:

1. Hold to Talk: Keep ⌥ Space pressed while speaking; release to finish.
2. Press to Toggle: Tap once to start and once to stop.

Customization: You can map any shortcut using ⌘, ⌥, ⌃, or F keys.

🧹 Intelligent Cleanup

BetterWispr removes the "noise" from your speech. For example: "and uh uh we can build um the export" $\rightarrow$ "We can build the export."

✍️ App-Specific Styling

Set a unique tone for different contexts:

• Formal: Preserves your exact wording.
• Casual: Removes the final period.
• Very Casual: Removes both final periods and capitalization (ideal for Messages).

---

🧠 Model Ecosystem

BetterWispr offers a variety of local speech models to balance speed and accuracy.

Built-in Options:

• Apple Speech: Native system integration.
• Parakeet TDT v3: Supports English, Japanese, and 25 European languages.
• Whisper Large v3 Turbo: High-accuracy via WhisperKit.

External API Options: If you prefer cloud-based or specialized AI, you can use your own API keys for:

• Sarvam AI
• Smallest AI
• OpenAI compatible APIs

---

📝 Advanced Features

🎙 Meeting Notetaker

BetterWispr can act as a meeting assistant by distinguishing between speakers:

• Me: Your microphone input.
• Them: Audio from other applications.

Once recorded, Ollama or Apple Intelligence can process these notes to generate:

• Executive Summaries
• Key Decisions
• Action Items

🎓 Personalized Learning

The app learns your specific vocabulary. If you frequently use technical terms like Postgres or TypeScript, you can add them as hints.

Pro Tip: If a word is transcribed incorrectly, fix it in the History tab or immediately after pasting, and BetterWispr will remember the correction for next time.

📊 Performance Insights

Track your dictation habits with a built-in dashboard:

• Total Volume: e.g., $4,218 \text{ words}$
• Speaking Pace: e.g., $137 \text{ wpm}$
• Efficiency: e.g., $312 \text{ words cleaned}$
• Consistency: Daily streaks.

---

🔒 Security & Transparency

BetterWispr is built for trust. Your data stays on your Mac.

• Local Storage: Both raw and corrected text are stored locally.
• Verified Updates: Updates are delivered via GitHub Releases through Sparkle.
• Signature Check: Installations only proceed if the EdDSA signature is verified.

If you are a developer, you can audit the code on GitHub or build the project from source:

make dev

---

⚖️ Comparison with Alternatives

---

💬 Community Feedback

"Just tried it out... I dictated about 5 different things in different contexts, it worked without a single error!" — GodelNumbering (HN)

"haha, product looks good btw goodluck :)" — Nikhil Bhima (X)

"sick stuff man" — Rahul Karajgikar (X)

---

❓ FAQ

How do I get started?

1. Install the app.
2. Follow the welcome guide to grant permissions.
3. Download your first local speech model.

Does it work offline? Yes, provided the local model and assets are installed.

How is the text inserted? Once you stop speaking, the text is transcribed and inserted into the active text field. If the app blocks pasting, BetterWispr automatically copies the text to your clipboard.

What is the transcription speed? Simulated tests show typing at $40 \text{ wpm}$ and speaking at $150 \text{ wpm}$, with minimal lag for transcription.