TypeWhisper 1.1 for macOS - Private Speech-to-Text

1 min read
speech-to-textmacosprivacywhisperdictationlocal-aitranscription
Originally from typewhisper.com
View source

My notes

Summary

TypeWhisper 1.1 is a macOS speech-to-text app that runs locally with no telemetry or subscription. It offers system-wide dictation via hotkey, six interchangeable speech engines (WhisperKit, Parakeet TDT v3, Apple SpeechAnalyzer plus Qwen3 ASR, Groq Whisper, OpenAI Whisper as add-ons), per-app profiles, file transcription with SRT/WebVTT export, and a local HTTP API for automation.

Key Insight

  • Engine choice matters more than brand. Parakeet TDT v3 (NVIDIA) is claimed 5x faster than Whisper for 25 European languages - worth testing if you dictate mainly in English, Finnish, or Russian/Cyrillic-adjacent European languages.
  • Apple SpeechAnalyzer needs macOS 26+ and has zero setup (OS manages models). Trade-off: only ~40 languages and “good” accuracy vs WhisperKit’s “excellent” at 99+ languages.
  • Per-app profiles are the real differentiator over built-in macOS dictation. Automatically switch language/engine/behavior when jumping between Slack (English), Telegram (mixed), and a Russian email - no manual toggling.
  • Local HTTP API + CLI + plugins make it scriptable. Transcription can be piped into existing automations instead of just typing into the focused window.
  • File transcription with subtitle export (SRT/WebVTT) makes this a dual-purpose tool: live dictation plus batch processing of meeting recordings or wearable transcripts.
  • One-time paid app rather than subscription is the stated anti-pattern to cloud-based competitors (Superwhisper, Whispering, etc.).