On-Device STT Digest · Mar 19 Daily Digest
New On-Device Mac Dictation Apps
- CursorTalk: CursorTalk is a native menu bar dictation app for macOS that runs fully on-device, supports...

Created by Yohei Kawano
On-device speech-to-text startup news, benchmarks, and engineering updates
Explore the latest content tracked by On-Device STT Digest
Ultra-simple engineering for real-time voice agents in apps:
useRealtimeChatPrivacy-first local STT accelerating on Mac, with system-wide integration and custom features proving mobile-ready.
BridgeVoice from BridgeMind launches for vibe coding—push-to-talk voice-to-text into Claude Code on Windows, Mac, Linux via Pro plan.
Key on-device...
Cacatua debuts on iOS with practical dictation features:
Ertiqah debuts Contextli, spotlighting privacy in voice dictation:
Privacy-first engineering win: Fully local voice typing deploys to any computer without apps or cloud—just Raspberry Pi, mic, button.
Bandicam for Mac launches AI Video-to-Text feature, enabling on-device transcription of screen recordings without external services.
Key engineering...
OpenWhispr delivers voice-to-text dictation with local/cloud speech recognition, AI agents, meeting transcription, notes, and privacy-first design across platforms. Ideal reference for seamless hybrid STT in mobile/desktop productivity apps.
Key engineering hurdles for real-time STT on mobile:
Fusion Scribe desktop app streamlines batch audio/video transcription locally with Whisper:
Plaud.ai boosts recording workflows with AI speech enhancement now on Web—improving voice clarity for clearer playback—and file search to query across all files.
Speakmac offers a subscription-free alternative to Willow Voice for Mac dictation:
Claude Code introduces new /voice dictation command, pitted against Wispr Flow (used for months by tester) in 4:33 YouTube head-to-head for coding accuracy. Key real-world benchmark incoming.
Indie dev launches floating push-to-talk overlay for Android dictation over any app:
Moonshine models lead for smallest on-device deployments in latest ASR rankings, with training recommendations by model size and hardware. Trelis eval uses AI-terms dataset and semi-private tiers.
Free local STT options using Whisper for offline dictation: