View only. Sign in to run this demo.
Audio → Structured Data
Turn a recording into structured data: it is transcribed, and the fields you describe are extracted from the text
How to use Audio → Structured Data
- 1Add a recordingUpload an audio or video file, or start from a ready-made recording with the task that suits it.
- 2Say what to extractDescribe the fields in plain words, including lists. The task decides the shape of the result.
- 3Choose how to listenLanguage, speaker detection, hints and error fixing: the same options as in the Transcriber.
- 4Check the resultRead the data, then the transcript it came from, and the schema that was made for it.
- Three steps run in turn: the recording is transcribed, a schema is made from your task, and the data is extracted from the text. If the transcript is wrong, the data is wrong, so read it.
- Change the task and a new schema is made. Nothing is saved: the recording and the result stay in this session.
Start from a ready-made recording
This task has a ready-made schema, so only the extraction runs. Change the text and a new schema is generated.
Reads the speech as English, with the general large model of the HH server.
Tells who speaks when, so that the transcript names the speakers. It works with the HH server only.
Sent to the model with the recording, so that it spells names and terms the way you wrote them.
A second pass corrects the text before the data is extracted from it. You can review every change in the transcript afterwards.
A PDF with the extracted data, to download after the run.
Add a recording first.
No structured data yet
Add a recording and click Extract data. The data appears here, with the transcript it came from.
