Score spoken practice for accuracy, pauses and filler words
Transcribe a learner's or presenter's speech and have Jev score each word or the delivery, so a coach sees where to focus.
The use case
Language tutors, speaking coaches and people rehearsing a pitch need feedback on many short practice runs, and a coach cannot listen to all of them. In these builds streaming speech recognition transcribes as the person talks; Jev then judges each word against the reading text, or scores the run on pauses, filler words, repetition and clarity. The learner gets a marked-up transcript and scores straight away. A tutor reviews the trend across sessions and listens to a few recordings, since recognition errors on accents can look like mistakes by the speaker.
Example workflow
- Record or stream audio in the browser
- Transcribe with a streaming recogniser such as Deepgram or Whisper
- Ask Jev a per-word Noul check against the target text, or Score questions on delivery
- Show a marked-up transcript and scores to the learner
- A tutor reviews trends and spot-checks recordings
Tools
- Jev (TypeSafe AI)
- Whisper
- Deepgram
Risks and guardrails
- Recognition errors on accents get blamed on the speaker
- Voice recordings are personal data
- Scores can discourage beginners if shown without context
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- 大声读 (dasheng) by WquGuru on Made with Jev: madewithjev.com/builds/dasheng
- 大声读 (dasheng): original post or repo by WquGuru: github.com/wquguru/dasheng
- talkr by aniol on Made with Jev: madewithjev.com/builds/talkr
- talkr: original post or repo by aniol: x.com/0xaniol/status/2101076982373191927
- Deepgram (official page): deepgram.com (checked 2026-11-19)