Meeting Transcription API
Turn Google Meet, Microsoft Teams and Zoom calls into structured, speaker-labelled JSON. One capture API gives you timestamps, language detection, recordings, summaries and action items, without a separate speech-to-text integration.
In closed beta. Contact us to get access.
Meeting transcription API vs speech-to-text API
A speech-to-text API transcribes audio you already have. A meeting transcription API also joins the call, captures it and tells your backend when the final transcript is ready.
| What it handles | Speech-to-text API | Unipile meeting transcription API |
|---|---|---|
| Joining the call | You bring the audio file or stream |
From a meeting URL, a join_at time or a calendar event |
| Waiting rooms and timing | Your own meeting bot |
Handled behind one Notetaker resource |
| Meeting platform | Not known |
Inferred from the URL, same schema for Meet, Teams and Zoom |
| Final transcript ready | Your own pipeline |
notetaker.transcript webhook |
| Summary and action items | A separate integration |
Enabled in the same request |
Transcription is included in the capture rate, with no separate speech-to-text line item. See meeting bot API pricing.
What a meeting transcript looks like in JSON
One transcript object per notetaker, with speaker-labelled utterances, start and end times, the detected language and, when available, word timing and confidence.
curl --request GET \ "https://api.unipile.com/v2/$ACCOUNT_ID/notetakers/$NOTETAKER_ID/transcript" \ --header "X-API-KEY: $UNIPILE_API_KEY"
{
"object": "NotetakerTranscript",
"notetaker_id": "nt_01k00000000000000000000000",
"status": "available",
"utterances": [
{
"id": "utt_001",
"speaker": "Alex Morgan",
"text": "Let's send the proposal tomorrow.",
"start_ms": 42800,
"end_ms": 45120,
"language": "en",
"confidence": 0.97,
"words": [
{
"word": "Let's",
"start_ms": 42800,
"end_ms": 43140,
"confidence": 0.98
}
]
}
]
}
Speaker-labelled utterances
Keep the conversation attached to the people who said it, when the selected source supplies speaker attribution.
Reliable timestamps
Every utterance has start and end times. Word-level timing and confidence are included when the processing source provides them.
Language detection
Autodetection is the default. Force a language only when your workflow already knows what the meeting will use. Same schema for Zoom meeting transcription, Google Meet and Microsoft Teams.
Real-time or post-meeting transcription?
Final transcripts first, live delivery only when the use case needs it.
Final transcript, the production path
The meeting ends, processing completes and your backend retrieves the result. The dependable fit for CRM notes, ATS interviews, coaching, support reviews and searchable archives.
Live delivery, when you opt in
Latency, completeness and provider support differ from final transcription. It stays experimental and disabled by default, so post-meeting products do not inherit streaming complexity.
A webhook that says it is ready
The notetaker.transcript webhook is a readiness notification. Fetch the resource once its status is available, so your database always holds the latest version. Events are covered in the Notetaker API overview.
From transcript to CRM, ATS or coaching workflow
Attach every final transcript to the record your users already work in, with the Notetaker API.
Create or schedule
Send a meeting URL now, add join_at for later, or use automatic meeting recording to create the notetaker from a calendar event.
Wait for availability
Listen for notetaker.transcript. Status and history events drive clear progress and error states in your UI.
Fetch and enrich
Retrieve the transcript, summary and action items, then attach them to the CRM, ATS or support object already in your product.
Continue exploring the Notetaker API
Go further with recording, scheduling and platform guides.
Meeting transcription API: frequently asked questions
Speakers, languages, formats and delivery, answered for developers.
A meeting transcription API joins an online meeting, captures it and returns a structured transcript to your application. Unlike an upload-based speech-to-text API, it also handles the meeting itself: joining at the right time, waiting rooms, platform detection, processing and a notification when the final transcript is ready.
Yes. Unipile sends a notetaker to Google Meet, Microsoft Teams or Zoom from a meeting URL, a scheduled join_at time or an event on a connected Google or Outlook calendar. The platform is inferred from the meeting URL.
Yes. Transcript utterances include speaker attribution when the selected capture and processing source supplies it, and every utterance keeps its start and end times.
Batch, or post-meeting, transcription returns the final transcript once the call has ended and processing is complete. Real-time delivery sends text during the call, with different latency and completeness. With Unipile, the final transcript is the production path, while live delivery is experimental and disabled by default.
Language coverage depends on the provider. With Unipile, automatic language detection is the default, so you do not need to set the spoken language. You can force a language code when your workflow already knows the meeting language.
Transcripts are delivered as JSON: one transcript object per notetaker, with speaker-labelled utterances, start and end times in milliseconds, the detected language and, when available, word-level timing and confidence. A webhook tells your backend when the transcript is ready to fetch.
Still have questions? Our team is here to help.
Talk to an expert