AI glossary · Beginner · Speech & audio
Speech recognition
Speech recognition (ASR, Automatic Speech Recognition, also called STT) is a technology that listens to what a person says and transcribes it into text.
About this term
- Level
- Beginner · Core ideas for first-time readers
- Category
- Basics
- Fields
- Speech & audio
- Added
- 2026-10-08
How it works
Sound captured by a microphone is first converted into image-like data showing how strong each pitch of sound is over time. A deep learning model looks at this data and estimates which sounds, characters or words it corresponds to. At this stage, it also uses language knowledge about which words tend to follow which, to choose the option that fits the context among words that sound alike. Today, a single large model that outputs text directly from sound (the end-to-end approach) is the mainstream, and some transcribe multiple languages with one model.
Why it matters
Speech recognition is the entry point for almost every service that needs to turn speech into data, such as voice assistants, automatic meeting notes, video subtitles and call center conversation analysis. The transcribed text can be used directly by LLMs for summarizing, translating and answering, making it, together with TTS, a core component of "AI you talk to." It is also important for accessibility, for example in situations where it is hard to use your hands or for real-time captions for deaf and hard-of-hearing people.
Things to keep in mind
Accuracy tends to drop with background noise, multiple people talking over each other, dialects and accents, and technical terms and proper nouns. Some models show hallucinations, inventing and transcribing words that were never actually said during near-silent segments, so it is best to have a person check important records. Also, voice is personal information, so whether recordings are sent to a server or processed on the device (on-device AI) is worth considering when choosing a service.
Example
The "automatic meeting notes" feature in meeting recording apps and video conferencing services works by using speech recognition to transcribe the conversation, after which an LLM summarizes the text and extracts a to-do list. Voice input keyboards on smartphones and automatic subtitles on video platforms such as YouTube are also typical examples of speech recognition in use.