Hosting Whisper model on Vultr Machine

[shikhar.crpf]

Hi,

I am using this endpoint https://api.openai.com/v1/audio/transcriptions , model- whisper-1 to transcribe audio files in n8n. but there is a 25 Mb limit which is making things difficult.

My question is if i host whisper open source model somewhere such as on Vultr Machine, will the limit be removed?

I don’t want to go through those steps of chunking the audio in order to manage with the 25 Mb limit so thought of hosting the model on a server.

[vb]

Hi and welcome back to the community!

There is another way to work around the 25MB limitation, by re-encoding the audio files.

Here is a post from @_j with a quick overview for you to try.

Whisper API, increase file limit >25 MB

It is possible to increase the limit to hours by re-encoding the audio. As the primary purpose of the service is transcription, you can use voice codec and bitrate. For example, a command to get exactly what you want. ffmpeg -i audio.mp3 -vn -map_metadata -1 -ac 1 -c:a libopus -b:a 12k -application voip audio.ogg Opus is one of the highest quality audio encoders at low bitrates, and is supported by Whisper in ogg container. [image] Silence detection: also a useful tool.