How to integrate Daily meeting with other live transcript service?

ronyu Member

Since Deepgram's live transcript quality doesn't meet our needs, we're exploring integrating Daily with alternative live transcription services. I couldn't find relevant information in the Daily documentation or guides. Does anyone have experience or ideas on how to approach this? Thanks.



  • mark_at_daily
    mark_at_daily Community Manager, Dailynista admin
    edited March 10

    We're sorry to hear that Deepgram isn't meeting your needs. Do you have a different transcription service that you're planning to use? In general, live transcription lacks the accuracy of post meeting processing. But, we've compared Deepgram to other services and have found it to be fast and relatively accurate given the real-time nature of the transcription.

    At the moment, we don't support other integrated options. But, daily-python has the ability to join a meeting and receive audio tracks from either individual participants or the entire meeting. Those tracks can then be forwarded on to a transcription service of your choosing. It's clearly more work than using Deepgram, but it is an option that will allow you to use an alternative service for live transcription.

    Hope this helps!

  • ronyu
    ronyu Member

    Hi Mark,

    Thank you for the quick reply. The other service we are considering are Rev.AI and AWS Transcribe.

    We are not happy with the accuracy and the speed of the live transcript. If you don't mind, would you share the preferred transcription configuration that you find provides the best quality of the transcription?

    Thank you for providing the link to the daily-python guide, it's gonna serve as a good starting point for the custom integration.


  • kwindla
    kwindla Community Manager, Dailynista admin

    cc @chad for current best practices for fast live transcription.

    Can you tell us a little bit more about your use case? For example, if you want sub-second transcription events that's possible with Deepgram as "interim results". But if getting transcription events every 2-4 seconds is fine, that's our current settings default and you'll get "final results" that are more accurate than interim results.

    Also, what languages do you need support for, and is there specialized vocabulary in the domain you're working in?

    As Mark said, we've done a lot of testing and Deepgram's word error rate is lower than everything else we've tested, for live transcription.