breaking issue with the endpoint /v1/listen?model=nova-2 after ±8 minutes transcribing audio samples always stops

[enimatek-nl]

After feeding 2-10 audio frames per minute to the v1/listen endpoint with model nova-2 we always get correct replies but after a while (±8minutes) we get confidence:0 and "" empty transcriptions. Example of such a reply:

{ "metadata": { "transaction_key": "deprecated", "request_id": "bbed2ddf-46e6-4d74-bfa9-3c1bf4e57242", "sha256": "f5c408138c3255a09c52804931731109344e0294319ad765d9793fd7b4dd8343", "created": "2025-02-06T10:12:08.247Z", "duration": 9.077375, "channels": 1, "models": [ "65c2022f-3683-4be7-a405-7a4f425434dc" ], "model_info": { "65c2022f-3683-4be7-a405-7a4f425434dc": { "name": "2-general-nova", "version": "2023-12-21.17729", "arch": "nova-2" } } }, "results": { "channels": [ { "alternatives": [ { "transcript": "", "confidence": 0, "words": [] } ] } ] } }

We do get charged (dashboard shows logs with 200 reply) and the seconds do match the audio sample.
We tried using 'bun' and 'nodejs', implementing different 'fetch' libraries and also using the official deepgram sdk.

They all end up with the same issue, after a few minutes we pay, but don't get any transcript back.

Is this a know limitation on our account or bug etc.?

[deepgram-community[bot]]

Thanks for asking your question. Please be sure to reply with as much detail as possible so the community can assist you efficiently.
_{Consider joining our Discord community for more opportunity to engage with your fellow Deepgram users. You can earn points which can be redeemed for cool stuff by being active in our communities!}

[deepgram-community[bot]]

Hey there! It looks like you haven't connected your GitHub account to your Deepgram account. You can do this at https://community.deepgram.com - being verified through this process will allow our team to help you in a much more streamlined fashion.

[enimatek-nl]

that link doesnt work, clicking continue returns me to the homepage of deepgram.
but our project id is:
c703bbcb-4469-4f6c-a26b-61fe02597bf6

[deepgram-community[bot]]

This request didn't have any issues in our logs. But it sounds like when you exceed total durations of 8 mins you see this problem.

it also looks like you are sending us the nl audio .

0:{
"name":"nova-2-general"
"version":"2023-12-21.17729"
"uuid":"65c2022f-3683-4be7-a405-7a4f425434dc"
"languages":[
0:"nl-BE"
1:"nl"
2:"nl-NL"
]
"tags":[]
"mode":"batch"
"formatted":false
"latest":true
}
]```
If you're sending small chunks (2–10 frames per minute), Deepgram may expect a more continuous audio stream.

Some models may behave unexpectedly with intermittent input.

As a test: consider sending *larger, more continuous chunks* instead of very sparse frames.

> ![avatar](https://avatars.slack-edge.com/2023-02-20/4855338233600_fad9174023ee3f640ce1_32.png) This message was sent by John Vajda from Deepgram, via our community automation.

[enimatek-nl]

Ok thanks for the check on the server side, we also found out that after a longer period suddenly fetch rolls in with the actual data fields filled, looks like after a few minutes the http calls start to yield intermittent results, we will try the callback approach which is maybe a better solution for intermittent input.