AINewsnow

Putting the punctuation back into Japanese ASR transcripts (F1 0.89 vs Whisper's own 0.61)

TL;DR — We released ja-punct-restore-v1, a small model that restores 、。?! into punctuation-free Japanese text, one prediction per character. On a real ASR benchmark (650 FLEURS Japanese utterances transcribed by Whisper, punctuation stripped, then restored) it scores punctuation F1 0.8881, against…

Read the full story at DEV Community — Machine Learning ↗

Timeline · 1 report

  1. 2026-11-19 06:02 · DEV Community — Machine Learning
    Putting the punctuation back into Japanese ASR transcripts (F1 0.89 vs Whisper's own 0.61)

More stories

  1. VibeVoice: Microsoft's Free Open-Source Voice AI Transcribes a Full Hour and Talks Back in Real Time — DEV Community — AI
  2. How to get TikTok transcripts in bulk: use the captions TikTok already made — DEV Community — AI
  3. Podcast transcripts from an RSS feed in Python, billed per minute — DEV Community — AI
  4. Local AI on a PC: what actually works (Whisper, translation, LLMs) — DEV Community — Machine Learning
  5. OpenAI shuts down whisper-1 and its other transcription models on 26 February 2027: a migration checklist — DEV Community — AI
  6. Whisper turbo vs. medium on Arm and CUDA: The Faster Model Still Flipped Between CPU and GPU — DEV Community — Machine Learning
  7. Offline live subtitles with Whisper and M2M100: how it works and what it costs — DEV Community — Machine Learning
  8. Whisper on Windows with nothing to install: how a desktop app ships whisper.cpp — DEV Community — AI

Read the Daily Brief: the day's most important AI stories →