1. Counting words
English text is segmented with Intl.Segmenter using word granularity when the browser supports it. A tested Unicode-aware fallback handles older environments.
- Contractions, possessives, hyphenated compounds, numbers, URLs, email addresses, and dotted abbreviations each follow explicit tested rules.
- Emoji and punctuation without a word are not counted.
[pause 2s]and[slide]are timing controls, not spoken words.- Input is limited to 500,000 characters to protect browser responsiveness.
The segmentation behavior follows the standardized ECMA-402 Intl.Segmenter specification.
2. Base formula
For example, 1,000 words at 130 WPM takes 461.54 seconds before modeled pauses, displayed as 7 minutes 42 seconds. Calculations retain decimals internally and round only for display.
3. Pace presets are planning choices
| Mode | Preset | WPM | Intended use |
|---|---|---|---|
| Speaking | Deliberate | 100 | Emphasis, ceremony, unfamiliar material |
| Speaking | Presentation | 130 | Default planning baseline |
| Speaking | Voice-over / conversational | 150 | Prepared narration or fluent delivery |
| Speaking | Fast | 160 | Dense, practiced delivery |
| Reading | Careful | 180 | Dense or unfamiliar material |
| Reading | Average | 238 | English non-fiction baseline |
| Reading | Brisk | 300 | Light, familiar material |
The 238 WPM reading default follows Marc Brysbaert’s 2019 meta-analysis estimate for adult English non-fiction silent reading. The other values are transparent product presets around plausible use cases, not universal population averages.
Speech rate varies with genre, speaker, preparation, and measurement method. Tauroza and Allison’s research is included as evidence that a single standard speech rate should be treated cautiously—not as proof that every presentation should use 130 WPM.
4. Natural pacing model
When Natural pacing is enabled, the tool adds deterministic heuristic weights. These weights are product modeling choices and are not presented as clinical or linguistic constants.
| Signal | Added time |
|---|---|
[pause Ns] | User-entered seconds, capped at 30 per marker |
[slide] | 2 seconds |
| Blank paragraph break | 2 seconds |
| Single line break | 1 second |
| Sentence punctuation | 0.5 seconds |
| Comma, semicolon, colon | 0.25 seconds |
| Ellipsis or em dash | 0.75 seconds |
URLs, emails, abbreviations, and control markers are protected from accidental punctuation double-counting.
5. Planning range and calibration
An uncalibrated result shows ±10% around the modeled time. After a valid timed rehearsal is applied, the range narrows to ±5% because the estimate uses evidence from the current user rather than only a generic preset.
Audience questions, emotion, slide problems, difficult passages, and interruptions can still move the real result outside the range.
6. Target fit and section flags
The target difference is target seconds minus modeled seconds. A difference within two seconds is treated as fitting. Suggested word changes use the active WPM and are rounded to a whole word.
A section is marked Long when it exceeds 90 seconds or 25% of the full target. The tool selects the source paragraph when clicked, but never rewrites it with AI.
7. Known limitations
- The first release is designed and tested for English; other languages can have different segmentation and timing behavior.
- Reading difficulty, comprehension, note-taking, disability, age, and second-language use affect reading time.
- Emotion, audience response, demonstrations, media playback, and slide operation are not inferred from text.
- Auto-scroll follows the estimate; it does not listen to or track speech.
- Strict broadcast, examination, contractual, or event limits require a real rehearsal and an appropriate safety margin.
Sources
Change log
- August 20, 2026: Published word rules, formula, preset limitations, pause weights, ranges, calibration, sources, and known limitations.
- August 19, 2026: Initial calculation engine and tested planning assumptions.