Hi all!
We are currently mainly using Assembly's streaming api for speech-to-text but as Deepgram's pre-recorded transcription is so fast we consider switching to that. We seem to get better transcriptions if we submit the entire file rather than streaming it.
The usual latencies for pre-recorded is great and would be fine for our use case. However, sometimes (roughly 1 in 10 requests), the call takes >15s. This is not tolerable for this use case.
I wonder if you have an option to get dedicated capacity or priority handling for pre-recorded transcriptions so that we can stay <3s latency for a 1min audio. Please let me know who I should reach out to for such a partnership request.