Short answer
Speech-to-text (automatic speech recognition) converts audio into text for captions, notes, search, and voice interfaces.
Why it matters
Transcription quality drives meeting tools and accessibility. Errors concentrate on names, accents, and noisy rooms—so human correction still matters.
Diagram
Article
Privacy
Default to short retention; restrict transcript search ACLs.
Examples
Podcast captions
A producer batch-transcribes an interview, corrects proper nouns, and publishes captions with timestamps for accessibility.
Support call analytics
A contact center transcribes calls, redacts sensitive fields, and uses categorized transcripts to find common failure reasons—not raw audio alone.
FAQ
What is word error rate (WER)?
WER is a common ASR metric that counts substitutions, deletions, and insertions versus a reference transcript, divided by reference word count. Lower is better, but task usefulness can matter more than a single WER number.
Can STT handle multiple languages in one recording?
Some systems support automatic language detection or multilingual models, but performance varies. Separate channels or explicit language settings often improve results.
How is transcription different from meeting assistants?
Transcription focuses on converting speech to text. Meeting assistants add scheduling integrations, summaries, action items, and CRM features on top of STT.
Sources
-
Google Cloud Speech-to-Text docs
GoogleOfficial cloud ASR documentation covering models and features.
-
OpenAI speech-to-text guide
OpenAIOfficial guidance for speech-to-text style APIs.
-
NIST
NISTNIST has long published speech recognition evaluation methodologies used across research.