deepgram/deepgram-python-sdk

Inconsistency latency in real time STT with nova-2-phonecall

wilson97 opened this issue · 3 comments

Screenshot 2024-05-17 at 3 06 03 PM

Attached is a graph of the latencies for each speech chunk in live transcription with nova-2-phonecall (including both speech_final true and false). You can see most of the latencies are < 0.5s, which is great, but also sometimes there are spikes above 1s. Since we are using deepgram to make phone calls, anytime the STT latency goes from low to high, it results a lag on our end. Can you guys take a look at this? Is there anything I can do to make sure there is no more latencies above 1s? Thanks.

This is the same issue as posted in Discord here:
https://discord.com/channels/1108042150941294664/1241150521348329542

This isn't SDK related, but would be happy to discuss in Discord. I cross posted in both just so we don't have someone helping in both places.

Hi @dvonthenen just updated the thread in discord with some more information.

@wilson97 could you please share ur STT logic that you are using that is fluctating around 0.5 to the 1 i am getting the latency around 0.9 to 1.2 .could you please share the code snipets and the logic please ?