A speech disfluency is any disruption in the flow of spoken language that is caused by the speaker. Types of speech disfluencies include stuttering and hesitations, as well as the fillers people insert to avoid awkward pauses while they find their next words and perhaps ensure there is no opening to allow interruption.
A few categories of speech disfluencies:
Fillers – words or syllables inserted into speech such as “er,” “um,” “like,” “well,” “so” and “uh.” Fillers don’t add to the meaning of what is being said but are very common in human speech – according to some estimates, fillers make up as much as 20 percent of spoken language.
Hesitations – It’s less common for a speaker to simply pause than to insert a filler but people’s speech patterns are rarely regular.
Repeated words, syllables or sounds – Stuttering is one example of this, in which speakers tend to get hung up on the starting sound of a word, repeat it over and over and have difficulty getting past it.
Repairs – Speakers may, for example, mispronounce a word and repeat it with the correct pronunciation before moving on.
False starts – Speakers sometimes interrupt their own sentences, beginning a new subject before finishing the original thought.
Prolongations – These may be used to allow the speaker more time to formulate the rest of a sentence or may be used simply for the effect, as in: “Aaaaaaaaaannnnnd… I win!”
Blocks – In this case, people can’t produce the word they want.
Most people use speech disfluencies frequently and also have them happen inadvertently. AI technologies like natural language processing (NLP) systems require training in disfluencies. In voice-related AI applications disfluencies may be added to make the speech seem more human. Recent AI assistants, for example, have begun to adopt disfluencies to sound more natural to the people they interact with. Hesitations and filler words, in particular, are employed to make the AI sound less robotic.