Check Japanese narration with AI
Compare a Japanese script with recorded speech, listen to the disputed syllables and keep a review record before repairing individual sentences.
Check Japanese narration with AI before replacing a whole recording because one name sounds wrong. The japanese-tts-audit-skill compares a reading derived from the script with kana recognised independently from the audio. A local report marks likely differences and gives you a position from which to listen.
A flag might reflect the narrator, the pronunciation dictionary or the recogniser. Alternative readings can be valid. Listen before deciding, and record your judgement against that run. When a repair is appropriate, keep the display text and work on the affected sentence separately.
What it does
- Check a brand or character name against its approved reading.
- Replay a short passage within a long narration.
- Distinguish a likely dictionary mismatch from an audible error.
- Keep the reviewer’s decisions alongside the corresponding findings.
- Prepare one replacement sentence without regenerating approved audio.
Install and use
- Install the maker’s Python CLI and local recognition dependencies, including the initial model download. Keep the original Japanese script and audio.
- Generate the report in a new folder. Open the local browser timeline and listen to the marked cases.
- Record the decision. If you authorise regeneration, use the existing speech service, model and voice where available.
FAQ
Source
Made by chitoku. Original skill: japanese-tts-audit-skill, published on GitHub under the MIT licence.
View on GitHub