How Speaking Coach measures filler words, pace, pauses and pitch
Updated 13 October 2026 · Speaking Coach guides · 6 sources
A practice tool, not therapy. These guides and Speaking Coach help with everyday speaking habits. App numbers are estimates from automatic transcription of English speech and can be wrong. If you stammer or stutter, or have voice problems, a speech-language professional is the right person to see.
- Speaking Coach analyses one recorded English take at a time, after you stop; it gives no live feedback.
- Filler words and pace come from automatic speech-to-text, so they are estimates. 1
- Pauses and pitch are measured on your iPhone from the audio itself.
- The pace and pitch hints are the app's own rules of thumb, not research norms.
- It is a practice tool, not speech therapy; people who stutter should see a speech-language professional. 2,3
This page explains, in plain terms, what Speaking Coach measures and how, so you know what the numbers mean and where they can be wrong. It is written by the team behind the app and describes the current version.
How a session works
- Choose a scenario: Freestyle, Interview answer, Presentation open or Pitch.
- Tap to record. The screen suggests 60 to 90 seconds; a take stops automatically at 3 minutes.
- Tap to finish. Analysis starts after you stop. Nothing is analysed while you speak.
- Read your report, which is also saved to your history on the phone.
Before any audio leaves your phone, the app asks for your permission. See where your recordings go for the details.
Filler words
Your take is converted to text by a speech-to-text service set to keep filler sounds, which are normally removed for readability 1. From those words, the app's server counts:
- Filled pauses: um, uh, er, ah, hmm, mhmm, uh-huh, uh-uh, mm-mm, nuh-uh.
- "Like", unless the word just before it is on a short list that usually makes it a verb or comparison (I, you, we, they, he, she, would, really, don't, didn't, feel, look, sound, seem, just, something, anything, nothing, much, more, and their past or -s forms where relevant). See how this rule works.
- "You know", unless it follows "do" or "did", or carries a question mark.
It does not count so, basically, actually, literally, I mean, well, right, kind of, sort of, repeated words or restarts. The report shows the filler count, the filler rate (fillers as a share of all words), and a list of when you hesitated, in seconds from the start.
Words per minute
Words per minute is the number of transcribed words, fillers included, divided by the time from your first word to your last. Silence before you start and after you finish is not counted. The hint uses the app's own bands: under 110 wpm "a little slow", 110 to 160 "comfortable listening pace", over 160 "fast". These are practice rules of thumb, not research standards; measured conversation is often faster 4. More in speaking rate.
Pauses
Pauses are measured on your iPhone from the loudness of the audio. A silent gap of at least 0.35 seconds between stretches of speech counts as a pause; silence before your first sound and after your last is ignored. The report shows the number of pauses and the longest. Background noise can hide pauses, and a very quiet voice can create extra ones.
Pitch variation
Pitch is also estimated on the phone. The app tracks the pitch of your voice through the take and reports how widely it varies, as the spread of the middle half of values in semitones. Under 2 semitones gets the hint "fairly flat"; otherwise "good expressive range". That threshold is the app's rule of thumb. If there is too little clear voice in the take, pauses and pitch show "Not measured".
The coaching note
When available, a two or three sentence note is written by an AI language model. It receives only the numbers (duration, words per minute, filler count and filler rate), never your audio or the words you said. The server checks that every number in the note matches a measured value and discards the note if not. If the note is missing, the metrics are still shown.
Limits to keep in mind
- English only. The filler list and speech-to-text are set up for English, and the app's interface is in English.
- Transcription errors. Automatic speech recognition works less well for some voices and accents 5, so counts are estimates.
- Not for live use. It analyses a recorded practice take. It does not listen in on meetings or calls, gives no alerts while you speak, and does not analyse video or body language.
- No transcript or replay. The report shows numbers and times, not the text, and the audio file is deleted from the phone after a successful analysis.
- Not therapy. It is not a speech therapy tool or a treatment for stuttering. Stuttering is different from everyday fillers and is diagnosed and treated by speech-language professionals 6,2.
Price
Your first analysis is free. After that, analyses need a subscription, which comes with a free trial; the app shows the current terms before you subscribe.
Common questions
Does Speaking Coach give feedback as I speak?
No. You record a take, and it is analysed after you stop. There are no alerts during the take.
Does Speaking Coach count "so" or "basically"?
No. It counts filled pauses such as um and uh, filler "like" and "you know" only.
Why does the app show a different filler count than I heard?
Fillers are counted from automatic speech-to-text, which can miss or mishear words, and the "like" and "you know" rules are simple. 5 Your ears are the final check.
Does it work for languages other than English?
No. Speaking Coach is built for English speech only.
Sources and further reading
- Deepgram Docs. Filler Words (checked 13 Oct 2026). developers.deepgram.com
- American Speech-Language-Hearing Association (ASHA). Stuttering and cluttering (checked 13 Oct 2026). asha.org
- NHS. Stammering (checked 13 Oct 2026). nhs.uk
- Yuan J, Liberman M, Cieri C (2006). Towards an integrated understanding of speaking rate in conversation. Proceedings of Interspeech 2006, paper 1795. doi:10.21437/Interspeech.2006-204 isca-archive.org
- Koenecke A, Nam A, Lake E, et al. (2020). Racial disparities in automated speech recognition. Proceedings of the National Academy of Sciences 117(14), 7684–7689. doi:10.1073/pnas.1915768117 doi.org
- National Institute on Deafness and Other Communication Disorders (NIDCD). Stuttering (checked 13 Oct 2026). nidcd.nih.gov