Using Live Transcribe

Create and submit a Speech live transcribe session to transcribe a live audio stream to text.

The live transcribe feature provides real-time transcription, which enables you to send audio streams and receive text results in real time. Real-time transcription is required for many use cases across industries such as healthcare, call centers, and media. For example, physicians and nurses use medical dictation, which requires real-time capabilities and increases work efficiency. With the release of live transcribe, you can get accurate transcription in your application in less than a few seconds. You can use the embedded text window to try live transcribe or refer to the API document for information about how to programmatically integrate with the OCI real-time transcription service.

Use the following steps to create and submit a Speech live transcribe session to transcribe a live audio stream to text.

  1. Open the navigation menu  and select Analytics & AI. Under AI Services, select Speech.
  2. Select Live transcribe.
  3. Under Settings, enter the following information.
    • Compartment: Select the compartment you want to work in.
    • Model type: Select the speech model to use for transcription.
    • Model domain: Select the domain of the speech model to be used.
    • Language: Select the language to transcribe in.
    • Punctuation (Optional): Select to insert punctuation into the generated transcription, and then select one of the following options.
      • Automatic: Inserts punctuation automatically.
      • Spoken: Inserts punctuation only when verbally spoken.
    • Partial silence threshold (Oracle model only): Enter how long, in milliseconds, the service waits for more speech, after it stops detecting speech activity, before ending the speech recognition.
    • Final silence threshold (Oracle model only): Enter how many milliseconds of silence after a word is spoken that the service waits to end the session.
    • Partial result stability (Oracle model only): Select the amount of confidence required for the latest tokens before returning them as part of a new partial result.
    • Enable customizations (Oracle model only): Select this checkbox to customize the session, then enter the following information.
      • Customizations: Select the customization that you want. If needed, change compartments.
      • Override entities (Optional): Only customizations containing several entity lists can be overridden
  4. Select Start session and begin to speak.
  5. When you want to complete the transcription, select Stop session.
  6. To view the JSON file, select View JSON.
  7. To reset the session, select Reset.