Recorded takes: try it without signing in
Your turn
Sign in with Pollinations to record. You pay with your own Pollen; the default models are free-tier.
Sing it: see which notes were in tune
Play a short tune, sing it back in your own key, one “da” per note, and code measures every note in cents. No sign-in needed: the pitch reading runs in this tab.
How it works
Code scores, not a model
Your take goes to openai/whisper-large-v3 with no hint of the sentence. Code lines its transcript up with the sentence word by word: heard right, heard as something else, not heard, or extra.
The transcriber is tested too
A synthetic voice (openai/tts-1) reads the same sentence and the same transcriber writes it down. A word it gets wrong even from a clean reading is greyed out and not counted.
The pitch reader is tested too
For singing, the same pitch reader listens to a clean tone of the tune in your key mixed with the quiet part of your own take. A note it misreads even then is greyed out and not counted.
Pace and pauses from your audio
Words per minute and pauses come from the recording's own loudness in this browser. The coach model only reads that scorecard and may not invent mistakes it does not list.