Skip to main content

Transcribe with AWS Transcribe

AWS Transcribe uploads the session audio to S3, transcribes it, and returns each word with a time and a speaker. The result is stored in the same place as Attach speakers, so the Speakers tab, naming, and exports work as they do there. You don't enter an attendee count, and up to 30 speakers are separated.

Audio leaves your machine

The entire meeting audio is uploaded to S3, and Transcribe and S3 charges apply. Sogon deletes the uploaded audio when the job ends, whether it succeeds or fails. The feature is off by default.

Setup​

  1. Install the aws CLI: brew install awscli
  2. Sign in with the AWS profile you will use: aws sso login --profile <name>
  3. In Settings → STT → AWS Transcribe, turn on Allow transcription with AWS Transcribe.
  4. Choose the AWS profile and enter the S3 bucket and Region. The region must match the bucket.
  5. Check that the status under the profile shows Signed in.
StatusMeaning
Signed in — expires in 3 hoursReady to use.
The token expired but can be refreshedJust use it; the CLI refreshes the token. No new sign-in is needed.
Sign-in requiredRun the aws sso login command shown on screen in a terminal.
The AWS CLI is not installedDo step 1.

Transcribing​

  1. Expand the session in Settings → History → Meetings.
  2. In Transcript, press Transcribe with AWS.
  3. If the bucket you entered does not exist, Sogon says so and offers Create <bucket>. Check the name and press it; the bucket is created with public access blocked, default encryption, and a rule that deletes objects under sogon/ after 7 days.
  4. When it finishes, view the result in the Speakers tab.

Uploading a session that was already transcribed asks for confirmation first. The result is replaced and you are charged again. Names you gave speakers are kept.

Good to know​

  • A recording in several segments is merged and uploaded as one file, because separate uploads would number speakers differently in each part.
  • In sessions that recorded both the microphone and system audio, Sogon uploads only the system audio track.
  • When the transcription language is Auto-detect, Sogon sends the job as Korean.
  • If another account already uses the bucket name, Sogon asks you to choose another name.