Using search to get start and end time of clips for audio extraction | Deepgram | Page 1

runic shore Jul 27, 2023, 5:43 PM

#

Hi there,

I am a new user to Deepgram and have setup everything, where I have a 5 minute speech file which I then send an array of sentences to be searched for, using the search field in the transcription prerecorded api.

However, when I clip the track from the start and end time the end of the speech seems to be cut off sometimes. Is there a way to make this more accurate?

Thanks 🙂

ripe crypt Jul 27, 2023, 5:51 PM

#

Hey there @runic shore

This feels like a result of an inaccuracy in timestamps. What model are you using?

runic shore Jul 27, 2023, 5:55 PM

#

ripe crypt Hey there <@117396166362005508> This feels like a result of an inaccuracy in t...

hey 404, I am using the nova model

#

this is my current code for sending the request

ripe crypt Jul 27, 2023, 5:59 PM

#

Looks good to me! I think there is a known issue with the accuracy of word level timestamps on nova. Nova was created with much more accurate transcripts in mind.

You could either add a buffer to either end of the start and end times. That's all I can think to suggest.

We're working on new and more accurate models all the time. Would you be interested in joining a Beta of one of those when they're ready?

runic shore Jul 27, 2023, 6:02 PM

#

Ah I see, I was thinking of adding a buffer but just wanted to see if there was something I could add to maybe get it more accurate. Thanks for the help 🙂 and sure I would love to try out the new beta models whenever

#Using search to get start and end time of clips for audio extraction