Home/Glossary/Real-time transcription

What is real-time transcription?

Real-time transcription turns speech into text while people are still talking. Words appear within seconds of being spoken.

Batch transcription instead works on a finished recording.

Real-time transcription streams partial results, then final ones, as the audio arrives.

Meeting audio48 kHzStreaming STTDeepgram, AssemblyAIChunkssentence, word, rawYour webhookduring the call
Live transcription pushes text to you while the meeting is still running.

How live transcription works

Audio chunksTens of millisecondsOpen connectionUsually a WebSocketFirst guessA partial resultFinal phraseLocked in with contextAudio chunksTens of millisecondsOpen connectionUsually a WebSocketFirst guessA partial resultFinal phraseLocked in with context
From a chunk of audio to a final phrase.

Audio is captured in small chunks, often tens of milliseconds long.

Each chunk goes to a speech recognition engine over a connection that stays open, such as a WebSocket.

The engine sends a first guess almost at once. It then locks in each phrase once it has enough context.

This is why live captions sometimes rewrite themselves. The model fixes earlier words as the sentence ends.

In a meeting, a bot inside the call sends the live audio to the engine.

The text comes back to your app while people are still talking.

It arrives as a stream of segments, each with a timestamp and a speaker label.

Real-time vs batch transcription

Real-timeWhile people talkPartial results, then final onesA little accuracy traded for speedThe only option for live featuresBatch, or asyncAfter the meetingRuns on the whole recordingSees the full contextUsually a bit more accurateReal-timeWhile people talkPartial results, then final onesA little accuracy traded for speedThe only option for live featuresBatch, or asyncAfter the meetingRuns on the whole recordingSees the full contextUsually a bit more accurate
The same audio, two moments, two tradeoffs.

Batch transcription, also called async, runs on the whole recording after the meeting.

It sees the full context, so it is usually a bit more accurate. Real-time transcription trades a little accuracy for speed.

That matters whenever you need the text during the call, not after it.

Many products use both: a live stream during the call, then a corrected async pass for the record.

What live transcription enables

Live text opens up product features that run during the call:

  • Live captions and access for people who are deaf or hard of hearing
  • Real-time meeting agents that answer questions or fetch data mid-call
  • Live sales coaching, objection detection, and battle cards
  • Instant compliance checks on regulated calls
  • Running notes that are already drafted when the meeting ends

How to pick a live transcription provider

Latency matters most: time to the first partial result, and time to the final text.

Beyond that, check speaker labels on the live stream. Check punctuation and formatting quality, custom vocabulary for domain terms, and language coverage.

Ask too if the provider can deliver a higher-accuracy async transcript of the same call.

If you plan to build agents that reply during the meeting, measure the whole loop.

It runs from spoken word to model response. Transcription is only one part of that loop.

Real-time transcription in practice

MeetStream bots stream meeting audio over WebSocket, roughly 200ms behind the room. They deliver both real-time and async transcripts.

So one integration covers live agents and polished post-call notes on Zoom, Google Meet, and Microsoft Teams.

A streaming engine leaves no post-call transcript of its own. The post-call copy comes from transcribing the recording again, with speaker identification.

See the MeetStream docs for the streaming reference. See the meeting bot API page to go deeper.

FAQ

Frequently asked questions

How fast is real-time transcription?

Partial results usually appear within a few hundred milliseconds to a second of speech. Final text follows shortly after.

The full latency depends on the audio capture, the network and the recognition engine itself.

Is real-time transcription less accurate than batch?

Slightly, in most cases. Streaming models see little of what comes next. Batch models see the whole recording. The gap has narrowed.

Many teams run a live stream for speed, plus an async pass for the record.

Do I need a meeting bot for real-time transcription?

On platforms like Zoom, Google Meet, or Microsoft Teams, something has to be in the call to hear it.

A meeting bot joins as a participant and sends live audio to the transcription engine. Attendees install nothing.

Put the vocabulary to work

One API to join, record, stream, and transcribe meetings on Zoom, Google Meet, and Microsoft Teams.