top | item 47119221 Show HN: Mellon – Free local speech-to-text for Mac with phonetic corrections 2 points| xavpoon | 7 days ago |voice.mellon.chat 1 comment order hn newest xavpoon|7 days ago Built this because I got tired of paying $15/mo for Wispr Flow when all I needed was accurate dictation that runs locally.Mellon runs Whisper on-device — no cloud, no subscription, no data collection. What makes it different from other local Whisper wrappers:- Phonetic-based corrections: catches homophones and accent mistakes using phonetic matching, not just AI- Correction history: shows you exactly what was corrected, why, and which dictionary caught it- Screen context: reads what's on your screen to spell names and jargon correctly- Agent mode: highlight text and give voice commands to rewrite/proofread in place- 99 languages, medical dictionary (600k+ clinical terms)It's free. macOS 12+. No account needed — just download and go.Happy to answer questions about the architecture or how the phonetic correction engine works.
xavpoon|7 days ago Built this because I got tired of paying $15/mo for Wispr Flow when all I needed was accurate dictation that runs locally.Mellon runs Whisper on-device — no cloud, no subscription, no data collection. What makes it different from other local Whisper wrappers:- Phonetic-based corrections: catches homophones and accent mistakes using phonetic matching, not just AI- Correction history: shows you exactly what was corrected, why, and which dictionary caught it- Screen context: reads what's on your screen to spell names and jargon correctly- Agent mode: highlight text and give voice commands to rewrite/proofread in place- 99 languages, medical dictionary (600k+ clinical terms)It's free. macOS 12+. No account needed — just download and go.Happy to answer questions about the architecture or how the phonetic correction engine works.
xavpoon|7 days ago
Mellon runs Whisper on-device — no cloud, no subscription, no data collection. What makes it different from other local Whisper wrappers:
- Phonetic-based corrections: catches homophones and accent mistakes using phonetic matching, not just AI
- Correction history: shows you exactly what was corrected, why, and which dictionary caught it
- Screen context: reads what's on your screen to spell names and jargon correctly
- Agent mode: highlight text and give voice commands to rewrite/proofread in place
- 99 languages, medical dictionary (600k+ clinical terms)
It's free. macOS 12+. No account needed — just download and go.
Happy to answer questions about the architecture or how the phonetic correction engine works.