Choosing the Best Transcription Provider: AWS vs. AssemblyAI vs. Default

Hi Customer Support, I discovered that we can switch between transcription providers "Amazon AWS" and "AssemblyAI" and "Default" (but it's not clear to me what the default option is tbh...). Is there info on which provider would be more accurate / suitable in general?

👀1

4 comments

Peter Wooden
·
·
Hi Paul Pagel! I'm from the engineering team. "Default" means we will choose the provider that we believe is best for your data. If you are experiencing issues with transcription quality, try choosing an explicit provider, and then re-upload your data to re-transcribe it, and see if it makes a difference. The best provider for you will depend on your situation, but I would recommend AssemblyAI for English and the other languages listed in the "best" section of their documentation, and Amazon Transcribe for other languages.
👍1
Paul Pagel
·
·
That's very helpful Peter Wooden, thanks for the explanation :)
Paul Pagel
·
·
Hey Peter Wooden , coming back to your response I have a small addition: Recently we have been experiencing problems with the recognition of participants in the transcribed recordings. Participants are no longer recognised correctly. What happens is that all speakers are assigned to participant 1, and max. 2 participants are recognised. Also, turn taking in the transcript is not correctly recognised. Meaning, that if a participant 1 is interrupted by participant 2, transcription still "thinks" that participant 1 is speaking. The spoken language in which this has been observed is English. I have now tried various options, as it turns out that our default setting seems to be Assembly AI. As the same problem occurs here. I have temporarily switched from default to Amazon AWS. Now the participant recognition in the transcript works better. As this appears to be a bug, I have already written to support about it, as I don't think this is an intended behaviour, especially due to the spoken language is English.
Peter Wooden
·
·
Thanks for the feedback Paul Pagel! I would expect AssemblyAI to work better than that. If you are comfortable sending me via DM a sample audio clip which I can use to reproduce the issue, I can try troubleshooting it with AssemblyAI. If not, all good, and I'd recommend staying with the Amazon option for now. Thanks again!