Speech To Text intermittently missing sentences of text

[CyrusJia]

Recently we noticed that the Deepgram API responds with sentences of missing text from time to time. So the same set of inputs/params passed to the API could be perfect, but when used again minutes later, a couple minutes of audio would be missing.

Some more info in case it's helpful (same audio file & parameters):
BAD request_id: 3c7f40eb-0464-4564-a2c8-246a9c602bd7
GOOD request_id: 3c240fd7-530c-4139-8def-ea1735d5e0ff

Here are the settings I'm using: model=nova-3-medical&detect_language=true&paragraphs=true&punctuate=true&diarize=true&utterances=true&keyterm={a bunch of keyterms}

[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.

[deepgram-community[bot]]

It looks like we're missing some important information to help debug your issue. Would you mind providing us with the following details in a reply?

• The programming language you are working in (e.g. JavaScript, Python).
• The deepgram product you are using (e.g Speech to Text, Agent API)

[CyrusJia]

I'm able to reproduce this problem via curl directly as well as Python.
I was using speech to text (non realtime)

[deepgram-community[bot]]

I noticed both 3c240fd7-530c-4139-8def-ea1735d5e0ff & 3c7f40eb-0464-4564-a2c8-246a9c602bd7 had a good amount amount of keyterms, looks like medical terms. They all seem to be the same across requests.

So I don't see anything wrong with either request in terms of what you sent us.

Sometimes our models won't always get the transcript correct, as our models are non-deterministic. Though if you see this behavior consistently let us know. It can happen from time to time, and the complexity of the keyterms being used here could occasionally trip the models up.

This message was sent by John Vajda from Deepgram, via our community automation.

[CyrusJia]

Thanks John. Is there general guidance or a rule for how many total keyterms we should stay under to maintain consistency?

[deepgram-community[bot]]

👋 Key Terms are limited to 500 tokens per request; anything beyond that will return an error.

What “500 tokens” actually means in this context:
Deepgram uses the same idea of tokens that LLMs use:
• A “token” is a chunk of text (not a character, not exactly a word).
• Rough rule-of-thumb for English: ~4 characters ≈ 1 token, or ~75 tokens ≈ 100 words, but this can vary.
https://developers.deepgram.com/docs/keyterm#key-term-limits

This message was sent by John Vajda from Deepgram, via our community automation.