Does not respect maximum content length for audio transcription

[ohthehugemanatee]

Which version of integration_openai are you using?

3.5.0

Which version of Nextcloud are you using?

30.0.5

Which browser are you using? In case you are using the phone App, specify the Android or iOS version and device please.

N/A

Describe the Bug

When an audio file is submitted for transcription, OpenAiAPIService does not check the contentlength before submitting the API request. If the submitted audio file is too long for the API limits, it throws an API request error like this:

 A TaskProcessing core:audio2text task with id 8 failed with the following message: OpenAI's Whisper transcription failed with: API request error: 413: Maximum content size limit (26214400) exceeded (26384610 bytes read) 

Expected Behavior

The integration should detect the large content size and chunk the audio file for multi-part processing, or recompress with ffmpeg and the voice codec, etc.

At the very least it should throw an error or visible user message when the file is selected for processing and advise the user remediate it.

To Reproduce

Try to transcribe an audio file larger than 25MB with OpenAI API. I'm not sure if the maximum content size is adjustable in localAI but you could test it there too if you know the value.

[julien-nc]

Thanks for reporting this.
The problem is that the provider (integration_openai) can't prevent the task to be scheduled. It can't add restrictions on the input audio file.
The Assistant can add a restriction but does/should not know which provider is used.

So we can add a global restriction of 25 MB for the audio input even if some other providers might accept those "big" files. Wdyt @kyteinsky ?

[ohthehugemanatee]

Another option:

• An admin field for maximum audio file size, defaulting to 25M
• Use PHP-MP3 or similar to split files during task run.

Or consider using one of the PHP OpenAI libraries... since OpenAI API compatibility is a requirement for the backend anyway. I understand orhanerday's open-ai library handles large files automatically.