Support Streaming ?

[jays0606]

1. is there a way to receive the results realtime by streaming ?
   • currently , only the final transcript is received (is_final=True)

I want the fastest possible stt result. Any ideas to optimize ?

@RyanChimienti

[jays0606]

i found a solution.

setting enforceRealtime = False makes the result much faster.

the result is fetched almost real-time.

[RyanChimienti]

Hey @jays0606 , you're right, enforceRealtime should always be False unless you're load testing. A couple of other tips to reduce latency:

• In the Deepgram configuration in the sample contact flow, you can set dg_interim_results to true to see non-final results, which will come at a faster cadence (every 1 second instead of every 3-5 seconds).
• For the dg_callback, instead of passing an https endpoint and receiving the transcripts in POST requests, you can pass a wss endpoint and receive transcripts over a websocket instead. This will likely improve latency by a small amount.

[jays0606]

Thanks for the reply.

Also, is there a way i can receive only FromCustomer audio input ?? Currently it is by default FromCustomer, ToCustomer 2 channels. Is only receiving FromCustomer possible? Will this also improve latency?

Thank you @RyanChimienti

=> this i've also accomplished.

Just modify code to receive on FromCustomer audio, and change the deepgram api queryString to

queryString.append("encoding=linear16&sample_rate=8000");