Upload a recording. Rhetor returns a scored report on fillers, pacing, vocal variety, and structure — plus the three things to work on next.
It appears 14 times in the first two minutes — soften nowhere, sharpen everywhere.
Rhetor measures four things you can act on. Each one is scored, timestamped, and explained in plain language — not a wall of numbers.
Every "um", "uh", "like", "you know" — counted and timestamped to the second.
Words per minute over the whole talk, plus where your deliberate pauses landed.
Pitch range and energy — are you landing emphasis, or droning in one note?
Opening, signposting, and whether the close actually closes the talk.
No account, no install, no waiting room. Rhetor runs a transcription, an acoustic pass, and an evaluator-style synthesis in parallel.
An MP3, WAV, or M4A of any talk — up to twenty minutes.
Transcription, acoustic measurement, and an AI evaluator run together, not in sequence.
An evaluator-style report with rubric scores, a pitch-and-pace timeline, and your top three actions.
Scores tell you where you stand. The chart and the rewrites tell you what to change — at which second, in which sentence.
A two-line chart over the length of your talk: pitch in Hz, words per minute, and grey bands where you paused. Hover any peak to see what you were saying.
Side-by-side phrasing edits at the exact second they happened — with a one-line explanation of why the new version lands.
"I think this is kind of an important point, you know, that we should sort of pay attention to."
"This is the point. Pay attention to it."
Why.You're at the climax of the open. Every hedge tells the room the speaker doesn't believe it either — strip them all.
No. Audio is processed in memory and discarded as soon as the report is generated. We keep the report itself behind an unguessable URL so you can come back to it; you can delete it at any time.
English transcription is the most accurate today. We can analyze most major European languages, but the rubric prompts are tuned for English speech-coaching conventions for now.
Up to twenty minutes per upload, or 100 MB of audio — whichever you hit first. Most talks under fifteen minutes return a report in about three minutes.
A single adaptive-thinking Claude Sonnet call handles synthesis; Deepgram does transcription; Parselmouth and librosa handle the acoustic pass. Two-pass prompting keeps cost under a tenth of a cent per minute.
No. Upload, get a report, share the link. If you want history across devices, an optional account lets you keep them in one place — but it's never required to analyze a recording.