Books

Speech and audio AI concept

What is speech to text?

Reference

What is speech to text? It is technology that turns spoken audio into written words, usually as a transcript. The system estimates which words fit the sound and returns them as text. It can work on a recording or process audio as someone speaks. Some systems also add punctuation, which is a formatting feature rather than part of the speech itself.

Reviewed 2026-10-09

Spoken audio, written words

Speech to text converts human speech into readable written text. The technical term for this task is automatic speech recognition, often shortened to ASR. The result is called a transcript: a written version of the words the system recognized. Depending on the system, the input can be a saved audio file or audio arriving from a microphone. Speech recognition can support uses such as dictation and captions, where spoken words need to appear as text.

A transcript is shaped by the purpose of the task. A full verbatim transcript may record fillers, pauses, laughter, and other audible details. A clean verbatim transcript removes some unnecessary words while keeping the meaning, so it is easier to read. These are different transcription styles; neither changes what the original speaker said.

How a system turns sound into text

A speech recognition system processes audio and estimates a sequence of words that fits the sounds. One traditional description separates this into two parts: an acoustic model represents patterns in speech sounds, while a language model estimates which word sequences are likely. A recognizer uses these kinds of clues to produce text. Systems differ in how they build and combine those parts, so this is a useful outline rather than a description of every design.

The output can include more than recognized words. For example, a system may add punctuation or provide a confidence estimate for recognized words. These features are added to the transcript; they are not extra sounds that were spoken. Recognition can miss speech when the audio is very poor or when settings such as language or audio encoding do not match the recording.

One task among speech technologies

Automatic speech recognition and speech to text usually refer to the same basic task: turning spoken words into text. Text to speech does the reverse, turning written text into generated audio. Speaker diarization is a related feature that labels which speech segments belong to which speakers in a recording. A product can offer several of these capabilities together, but each produces a different kind of result.

A transcript answers, “What words did the system recognize?” It does not, by itself, turn text into a voice or provide speaker labels. Those outputs require different speech tasks. Knowing which result you need—text, generated audio, or speaker labels—helps distinguish transcription from its neighbors.

A transcript is an estimate

Speech-to-text output is a system’s recognition of speech, not a perfect copy guaranteed to include every word. Some transcription styles keep fillers and other audible details; clean verbatim styles may leave unnecessary words out. Punctuation can also be inferred and inserted by the system. As a result, two transcripts of the same audio can look different because they use different settings or transcription styles.

Poor-quality audio or mismatched recognition settings can lead to speech not being recognized. When a passage matters, compare uncertain words with the audio, especially if a name or uncommon term could have been recognized incorrectly. This is a practical check: a transcript is text produced from audio, and the recognition result may not capture the speech correctly.

Questions people ask

Is speech to text the same as automatic speech recognition?

Usually, yes. Automatic speech recognition is the technical term for processing human speech into written text; speech to text is a common name for that task.

How does AI transcription work?

It processes audio and estimates a word sequence that fits the sounds. Acoustic and language patterns can help a system recognize speech and produce the transcript.

Does speech to text identify who is speaking?

Its central task is to turn speech into text. Speaker diarization can add labels that distinguish speakers’ contributions in a recording; it is a separate output feature.

Is a speech-to-text transcript always word for word?

No. Full verbatim transcription can include fillers and pauses, while clean verbatim transcription removes some unnecessary words while keeping the meaning.

Is speech to text the same as text to speech?

No. Speech to text turns spoken audio into written text. Text to speech turns text into generated speech audio.

Learn more

What changed

Sources

Part of Speech and audio AI: Choose the speech task and measure its output

Questions and comments