AI Glossary / Natural language and speech / Beginner
Glottal stop
Definition
A glottal stop is a speech sound produced by a closure at the glottis, where the vocal folds come together and interrupt airflow. Its standard IPA symbol is [ʔ]. It is an articulatory event rather than simply an empty space in a recording, and its linguistic role depends on the language, variety, and context in which it occurs.
What it means
Stops can be formed at different places. A bilabial stop closes the lips, while a glottal stop creates its closure at the vocal folds. Describing the place helps distinguish sounds that can all involve a brief interruption of airflow.
In some systems of speech description, glottal stops are represented as separate sounds; in others, glottal activity may be described as part of a contextual realization. A phonetic symbol records an analysis of production, while a phoneme analysis asks how sound contrasts function in a particular language. Those are related but different questions.
How it works
During a glottal closure, airflow through the glottis is interrupted. The timing of closure and release, and its coordination with neighboring sounds, shape the acoustic result. A broad transcription [ʔ] does not by itself specify every detail of that timing or accompanying voice quality.
An annotator uses the recording and the corpus's conventions to decide whether and how to mark the event. Some speech resources use a symbol such as q for a glottal stop in their own transcription scheme. That does not make the ordinary written letter q universally equivalent to [ʔ].
Automatic analysis requires care because a low-energy interval can have several causes. Silence, an oral stop closure, a recording problem, and a glottal event are not interchangeable labels. A system should use appropriate contextual and acoustic evidence rather than assigning glottal stop to every quiet gap.
Three practical examples
1. Speech-corpus annotation: Reviewers mark glottal stops according to a documented phonetic convention, making the labels consistent across speakers and recordings.
2. Pronunciation modeling: A team represents a supported pronunciation variant in its speech model without assuming that the written word has changed.
3. Educational audio: A lesson compares a glottal closure with a lip closure, using clearly labeled examples to explain why place of articulation matters.
Think of it this way
Think of briefly closing a gate in an airflow path. The interruption happens at that gate rather than at an opening farther along the path.
The analogy is limited: real vocal folds interact dynamically with airflow and neighboring articulations. A glottal stop is not merely a pause button applied to a sound file, and it should not be inferred just because the waveform becomes quiet.
Worked example
Imagine a fictional annotated recording sampled at 16,000 samples per second. A reviewer marks an interval from 0.420 to 0.470 seconds as a glottal closure after examining the relevant evidence and annotation conventions.
The interval lasts 0.050 seconds, or 50 milliseconds. Under a half-open timing convention, its start and end correspond to sample indices 6,720 and 7,520, so it spans 800 samples. These values come from multiplying each timestamp by the sampling rate; they do not come from counting visible letters in a transcript.
Now suppose another 800-sample low-energy interval occurs elsewhere in the file. Its equal duration does not establish that it is another glottal closure. The event might have a different cause, and the label requires separate evidence.
This example illustrates timing conversion after a stipulated annotation. It does not prescribe 50 milliseconds as a typical or diagnostic duration, and no real audio was analyzed. It also shows why an exact numerical boundary can coexist with uncertainty about the phonetic interpretation: precise arithmetic cannot replace the evidence needed for the label.
Illustrative case study
Fictional case: A speech-recognition team reviews errors around brief interruptions in spoken words. An initial diagnostic script labels all low-energy spans as glottal stops, inflating the apparent frequency of the phenomenon.
Phonetics reviewers inspect representative segments and separate several causes of low energy. The team revises its annotations and records uncertain cases instead of forcing a single label. It then evaluates recognition errors using the corrected categories, avoiding a misleading explanation based on the first script's output.
Common misconceptions and mistakes
A glottal stop is not simply any pause. The term identifies a place and manner of articulation, while pauses can arise without that closure.
It is not the same as a voiced sound in general. Vocal-fold vibration and a complete glottal closure describe different behaviors, although speech can move between them in a short interval.
The IPA symbol [ʔ] should also not be confused with a question mark or with ordinary spelling. A transcription system's q notation, where used, needs explicit interpretation. Finally, a glottal stop's occurrence does not automatically indicate incorrect speech; language and variety determine its role.
Business use
For businesses using speech technology, careful glottal-stop handling can improve the quality of pronunciation resources and error analysis. The goal is to represent the speech varieties relevant to the application, not to impose one assumed pronunciation pattern on every speaker.
Keep annotation guidelines, audio quality, and reviewer uncertainty visible. Test whether adding a pronunciation variant actually improves the intended system. A label can help diagnose a problem, but neither its presence nor a numerical duration alone explains every recognition or synthesis error.
People also ask
What is the IPA symbol for a glottal stop?
It is [ʔ], a phonetic symbol distinct from ordinary punctuation.
Is every quiet gap a glottal stop?
No. A quiet interval has several possible causes, and identifying glottal closure requires suitable evidence.
Is a glottal stop always a separate phoneme?
No. Its phonological status depends on the language and analysis; a phonetic occurrence and a contrastive phoneme are different notions.
Check your understanding
At 16,000 samples per second, how many samples span a 50-millisecond interval, and does that count prove the interval is a glottal stop?
Answer: It spans 800 samples. The count establishes duration under the stated convention, but it does not establish the articulatory cause or justify the phonetic label.
Related terms
- Glottal
- Stop (consonant)
- Phoneme
- glottis
- phonetic transcription
- allophone
- speech annotation
Learn next
Learn next about stop closures and phonetic versus phonemic transcription. Inspect how a corpus defines its symbols and boundaries, then distinguish an annotated event from the acoustic cues used to identify it.
More in Natural language and speech
- Generics - Generics are statements that describe a kind or category in general without explicitly saying how many members have the stated property.
- Genetic fuzzy systems - Genetic fuzzy systems combine fuzzy-system representations with evolutionary search, commonly genetic algorithms, to learn or adjust parts of the system.
- Given-new - Given-new describes a distinction in how information is presented in discourse.
- Glyph - A glyph is a visual form used to represent written text or symbols.
- Gold labels - Gold labels are the reference annotations treated as the accepted answers for a defined machine-learning task or evaluation.
- Gorn address - A Gorn address identifies a node in an ordered rooted tree by recording the sequence of child choices needed to reach it from the root.
Sources
Jurafsky and Martin's phonetics chapter was read for glottal closure and corpus transcription examples: https://stanford.edu/~jurafsky/slp3/15.pdf . The International Phonetic Association's official chart was consulted for [ʔ]: https://www.internationalphoneticassociation.org/IPAcharts/IPA_charts_TI/IPA_charts_TI.html . The 50-millisecond interval and recognition-team case are original hypothetical examples, not observed measurements or claims about typical duration.