How FluentAI measures speaking, posture and clarity

Your speaking is transcribed as you talk. From that transcript FluentAI counts words, sentences and filler words — sounds and phrases such as “um”, “uh”, “like”, “يعني” and “الحين” — and divides them by the length of the session to report filler words per minute. Pauses of two seconds or more are measured from the timing of each recognised phrase.

Pace is reported in words per minute. The target band is 120 to 160 words per minute: below it, listeners can lose the thread; above it, articulation and clarity suffer. Pauses, pace and filler control trade against each other, which is why the report shows them together rather than as one vague “fluency” figure.

With the camera on, a pose model locates your shoulders, hips, head and wrists in each frame — about ten times a second. Because several people can appear in a room, FluentAI keeps following the person nearest the camera and centred in frame, so a colleague in the background never takes over the measurements.

Posture is scored from shoulder tilt and torso lean relative to your own body size, not from absolute pixels. Hand gestures are scored from the share of frames in which your hands are raised in the speaking zone, and if your hands never enter the frame the report says so instead of scoring gestures. Camera engagement is scored from how often your head is directed toward the lens. Body stability is scored from frame-to-frame displacement, which catches fidgeting. If no person is detected, the report says so instead of inventing a score.

Speech metrics measure delivery; they cannot judge whether an interview answer actually answered the question. That part is done by an AI coach working from your transcript and the mode you chose — presentation, interview, general conversation, or climate communication. It scores content clarity, vocabulary, grammar, confidence and structure, and writes the strengths, weaknesses, specific examples and next practice steps you see in the report.

Because the coach reads your real words, its advice can quote you. Each score in the report carries the reason behind it, so you can disagree with a judgement and still learn from the measurement underneath it.

Arabic and English behave differently as speaking material. Modes and prompts exist in both languages, filler detection uses a language-specific list, and the transcript is kept in the dialect you spoke — Gulf, Egyptian, Levantine or Modern Standard Arabic — rather than being silently normalised. Mixed Arabic and English sentences keep their English terms, which is how many speakers actually talk.

Camera frames are analysed locally; the images themselves are not stored or sent anywhere. Sessions, goals and scores are saved in this browser's local storage under an anonymous identifier, not an account. Clearing your browser data removes them permanently, and nothing syncs between devices — the trade for having no profile to leak.

Choose a mode and a prompt, allow the microphone, and speak for at least fifteen seconds and ten words. When you stop, FluentAI transcribes the recording, measures speech and posture, scores the session, and opens a report with your overall score, the breakdown, the evidence behind each score, and the recommended focus for next time.