Continued Live Transcription of Audio in Python

[sadmoody]

I've been using Deepgram's Python SDK to transcribe audio. I want a constant stream of transcription to be put through from Python, but it appears that the SDK doesn't really prompt with responses until you await finish.

I'm using pyaudio to listen to my input device, sending chunks of data as they get generated. The callbacks aren't triggered until I stop sending data and send the empty websocket message.

Is there a way to constantly send data in Python without needing to signal a stop periodically in order to get transcripts?

Thanks!

[geekchick]

Hi @sadmoody ! Thank you for your question, and we appreciate you! I'm Tonya, the Python Developer Advocate at Deepgram. To make sure I understand your question, are you asking if there's a way to continually receive data in python without terminating or stopping the program to get the transcripts?

[sadmoody]

Awesome work @geekchick 🎉

I'll get out of your hair in that case since you'll have more visibility over where the issue might be. My project is a little weird in its entirety, but I'd assume that this specific function wouldn't be too uncommon.

No need to apologise - glad you're on the case! Ping me if you come up with anything.

[geekchick]

Thank you for your understanding @sadmoody ! I'll continue working today on troubleshooting this issue. Will be in touch soon.

[geekchick]

Hi @sadmoody I didn't forget you! I'm back to troubleshooting this issue today.

[sadmoody]

Have you managed to make any progress @geekchick ?

I've almost run out of other things to do on my project!

[sadmoody]

I had a showerthought about this for a workaround which I'll try in the next couple of days. I'm suspecting that there's something in Deepgram's engine that doesn't like the raw signal being sent. If we process the wave input into a different format before sending it, that might work for now.

It adds a bit of overhead, but it gets me going again. This problem occurs when recording from a microphone but not when streaming the bbc audio for example. If we process the microphone audio into whatever format the bbc audio data is, that should work.

[geekchick]

Hi @sadmoody ! Thank you for your patience. I grabbed a code snippet from one of my colleagues here at Deepgram. Hitting they API directly, they were able to get live streaming working. I'm going to attached the python file pyaudio-streaming and the requirements.txt file below. I'm playing around with this code to figure out if I can adapt it to the Python SDK. Can you let me know if you were able to run the code sample?

requirements.txt
pyaudio-streaming.txt

[sadmoody]

Thanks so much - will check later tonight!

[sadmoody]

We're in business! Works great on my end and is blazingly fast! Will integrate this for now, and then move to the sdk when you get around to it. This isn't a blocker for me any more.

Thanks so much @geekchick! @michaeljolley, make sure you give "big ups" to Tonya and this mystery colleague at the next standup. You need to use the phrase "big ups" in the sentence. Not accepting any regional alternatives thank you.

[geekchick]

Hi @sadmoody ! I'm so happy to hear that worked for you! :)