Sync Your Speech
Record yourself reading a short script. Get a word-by-word pronunciation report in seconds, with the sounds you missed and how to fix them.
Pronunciation report
session_04.webm
Flagged words
- tech-nol-o-gytechnology
- in-no-vay-shuninnovation
- ex-per-i-enceexperience
Next step
Stress the second syllable of technology — it carries the meaning, not the first.
How a session runs
Three steps, one report
The whole loop is a recording and a model pass. Nothing to book, nothing to install.
- 01
Read the script out loud
A short paragraph appears on screen. Hit the mic and read it naturally, at your own pace.
- 02
Whisper transcribes every word
Your audio is sent to a Cloudflare Worker running Whisper. It turns speech into timed text in seconds.
- 03
Llama grades your pronunciation
A second model compares each word against correct pronunciation and writes specific fixes for the ones you missed.
Questions
Before you press record
A pronunciation tool for non-native English speakers aiming at remote work. You read a script, record yourself, and get a word-by-word report on what you mispronounced and how to fix it.
You record on the dashboard, the audio is transcribed by Whisper, then a Llama model grades each word against correct pronunciation and writes specific recommendations. The report lands on /analysis.
A separate Cloudflare Worker. Whisper handles transcription, a durable Workflow runs the Llama analysis, and the result is stored in KV keyed by the job id. The web app talks to it over a service binding, so there is no public endpoint for your audio.
No. It runs in the browser on any device with a microphone. The recorder uses MediaRecorder, so a recent Chrome, Edge, Firefox, or Safari is enough.
The audio is sent to the worker for transcription and discarded after the analysis is written. Only the structured report is kept in KV so you can view it on the analysis page.