Show HN: ZWhispr – Three speech models are better than one

Category: ai-ml

Tags: speech-recognition, macos, ensemble, dictation, privacy

Score: 5.3/10 (Innovation: 6, Technical: 5, Documentation: 4, Utility: 5)

ZWhispr is a macOS dictation app that runs multiple speech recognition models in parallel and intelligently merges their transcripts to reduce errors and latency. It optionally uses on-screen context (extracted locally) as vocabulary hints and types the final text into any app. The parallel ensemble approach is a clever way to improve accuracy and tail latency.

Target audience: macOS users, productivity-focused professionals, developers, writers

Repository: https://zwhispr.com/

View on Hacker News