Show HN: BetterWispr – Free, open-source dictation for Mac

Original Article ↗ Hacker News Discussion ↗

A new macOS app called BetterWispr offers free, open-source voice dictation that runs locally using models like Parakeet and Whisper, aiming to avoid subscriptions and cloud processing. Commenters compare it with a growing ecosystem of similar tools (Handy, VoiceInk, Fluidvoice, Apple’s built-in dictation, Google’s Eloquent, etc.), weighing trade-offs in accuracy, latency, UI design, and privacy. Many see high-quality local dictation as key to new workflows such as voice-driven coding and meeting note-taking, but note the space is crowded and that long-term value may come from features like diarization, better customization, and tighter integration with AI assistants.

Overview & Requirements

  • BetterWispr is a free, open-source macOS dictation app that runs speech recognition locally (Whisper, Parakeet, or Apple Speech).
  • Main UX: hold Option+Space to dictate into any active text field.
  • Focus areas: accuracy, speed, filler-word cleanup, custom vocabulary, app-specific styles, and optional meeting transcription/summaries.
  • Minimum requirement mentioned: macOS 14.

Comparisons to Other Dictation Tools

  • Frequently compared with Handy, VoiceInk, Fluidvoice, and various small CLI/GUI tools.
  • Some see Handy as simple, fast, and focused on streaming with a good model picker; others call it bloated with a large binary and poor accuracy for certain accents.
  • Some users recently switched from Handy to alternatives (Fluidvoice, VoiceInk) citing better UI or features.
  • Several people built or shared tiny or DIY alternatives (e.g., very small binaries, CLI tools) and voice-focused agent forks.

Models & Technical Choices

  • Parakeet v3 is widely praised for speed and accuracy on modern Macs; multiple commenters say it’s become their default over Whisper.
  • Others still prefer Whisper/OpenWhisper for accented speech.
  • BetterWispr explicitly uses Parakeet v3 by default and positions itself as improving both accuracy and developer experience.
  • Discussion of combining multiple models in parallel for higher accuracy, or using LLMs to post-process and clean dictation output.

Native macOS Dictation & Competition

  • Some argue macOS’s built-in dictation (Parakeet-based) is “good enough” and comparable to top third-party tools.
  • Others report noticeable latency, missing features like custom dictionaries and mic selection, and see room for third-party apps.
  • General sense that the dictation space is crowded, with frequent new clones.

Use Cases, Ergonomics, and Workflow

  • Several users now code, chat with LLMs, or write long-form text almost entirely by voice, citing:
    • Faster throughput than typing.
    • Better ergonomics for neck, back, and wrists.
    • Ability to “ramble” and let an LLM summarize and structure intentions.
  • Others find speaking tiring, socially awkward in offices, privacy-unfriendly, or simply slower than their typing.

Meeting Notes, Diarization, and Advanced Features

  • Some see the main unmet value not in pure dictation but in:
    • High-quality diarization (speaker separation).
    • Intelligent meeting summaries and note generation.
    • Integration with local LLMs to validate statements in real time or detect “bullshit.”
  • BetterWispr’s note-taking / meeting features are highlighted as aligned with this direction.

Feature Requests & Ideas

  • Requests include:
    • Windows/Linux support.
    • Wake-word activation and silence-based auto-stop.
    • Better OS/IDE control via voice (“open this PR in editor,” app switching).
    • Separate modes or buttons for dictation vs. voice commands to agents.
    • Real-time voice-clone output for recording with noisy environments.
    • Bluetooth remote, pedals, and rings mapped to dictation shortcuts.
    • Using LLMs (local or cloud) for advanced cleanup, including of existing non-dictated text.

Skepticism, Friction, and Naming Concerns

  • Some think most new dictation tools “suck” or regress compared to older solutions, especially for long-form transcription.
  • One blog referenced argues transcription quality overall has worsened, but others respond that users can still use simple system dictation without AI rewriting.
  • There is brief concern about the “Wispr” name and potential trademark risk; the situation is noted as unclear.
  • A claim that Bluetooth audio lag is “fixed” in BetterWispr is met with skepticism, with at least one commenter suspecting exaggeration.