Why I built it
Pronunciation products often present confidence-heavy scores without enough evidence that the score is valid or useful for a specific learner.
What exists
The family includes a substantial web interface, external phoneme-analysis service, realtime audio experiments, profiles, and reports; the public core flow and scoring validity remain incomplete.
How it works
A web application connects profile and session state to separate phoneme-analysis and realtime-coaching paths.
- Audio capture interfaces
- Phoneme-analysis integration
- Realtime coaching experiments
- Learner profile and report surfaces
My role
Designed and built the product surfaces, speech-analysis integration, realtime agent experiments, learner profiles, and reports.
What I checked
- repositorySubstantial web and service implementation audited
What stays bounded
- No objective scoring claim
- No proven learner improvement
- Synthetic audio and transcripts only in public evidence
What is missing
- Scoring validity, API security, persistence consistency, tests, and the complete core flow need work.
Next
Rebuild one narrow pronunciation exercise with human-reviewed labels.
Later ideas, not current features
- Choose one accent influence and audience
- Define one measurable speech outcome
- Create a human-labeled evaluation set