Presspeech vs VoiceInk
These are philosophical neighbors: both open source, both local-first, and both can run NVIDIA's Parakeet models through FluidAudio. VoiceInk is a configurable workspace with many models and AI enhancement. Presspeech is one fixed pipeline tuned for latency.
| Question | Presspeech | VoiceInk |
|---|---|---|
| Primary workflow | Hold hotkey, speak, release; local model transcription is followed by deterministic text handling and normally a paste after confirming the same window is still focused, with manual clipboard recovery when needed. | Hotkey dictation with optional AI enhancement, per-app Modes, and audio/video file transcription. |
| Engine path | Local Parakeet TDT v3 CoreML model on the Apple Neural Engine via FluidAudio. | Your choice: Parakeet V2/V3 or Parakeet Ultra (a post-trained V3) via FluidAudio, Whisper models via whisper.cpp, Apple Speech on macOS 26+, or optional cloud providers. |
| Privacy posture | No cloud transcription or AI path, account, or Presspeech-authored analytics; transcript history is memory-only. Model downloads and optional update checks use the network; see the version-specific network inventory for published-build token and library-telemetry caveats. | Local models are the default. Optional cloud transcription sends audio; optional cloud enhancement sends text and any context you enable. Transcript history is stored locally until you delete it or configure cleanup. |
| Price and license | Free and MIT licensed. | Free trial, then $25 one-time for one Mac, $39 for two, or $49 for three; GPL-3.0 source you can build yourself for free. |
| Footprint | 8.4 MB app, ~80 MB RAM idle, 0% CPU between dictations; published benchmark. | 48.7 MB v2.21 DMG; resource figures not published, 8 GB RAM recommended. |
| Configurability | Focused and deterministic: hotkey and trigger style, language pinning, local corrections and shortcuts, spoken formatting, and filler removal. | Extensive: model management, Modes with custom prompts and app triggers, a dictionary that can learn from corrections, and searchable history with optional audio replay. |
| Platforms | Apple Silicon Macs on macOS 14+; a separate Windows preview is available. | The prebuilt Mac app requires Apple Silicon; v2.21's README raises the minimum to macOS 15.0 (its site still lists 14.4+). A separate iPhone/iPad companion is in beta. |
Choose VoiceInk if you want one configurable tool that also handles AI cleanup, per-app behavior, saved history, and file transcription, and a one-time price fits. Choose Presspeech if you want a smaller fixed utility without cloud transcription or persistent transcript history, or need macOS 14 support. Both Mac apps can run Parakeet, so the larger difference is product surface and data-flow choice.
Sources: Presspeech source, Presspeech benchmarks; VoiceInk's workflow guide, model catalog, history documentation, and privacy policy (checked 22 September 2026); and the VoiceInk official site, iOS beta, v2.21 README, v2.21 release, and release assets (checked 22 October 2026).