Repository navigation
Replies: 2 comments
|
you are right, the web ui does not have the field. i checked the source and the music page posts to POST /v1/sound-generation. the request schema for that endpoint has no audio reference parameter at all (core/schema/elevenlabs.go). the handler also hardcodes nil for the source file when it calls the sound generation backend (core/http/endpoints/elevenlabs/soundgeneration.go). the backend protocol itself does support it through the src field in backend.proto. so the ui simply cannot reach cover mode today. the workaround that works now is the cli: local-ai sound-generation -b ace-step -m -i reference.wav -d 120 "your prompt" the -i flag is the input file to condition generation on and it is passed straight to the backend as the source file. there is also -f if you only want the first portion of the sample used. so yes to your first question, it is a ui and api gap. and your third question is the real fix: the ui needs a file upload field and the endpoint needs to pass the source file through instead of nil. |
|
Thank you so much @aashish254 |
Uh oh!
There was an error while loading. Please reload this page.
Hi everyone,
I am currently exploring the audio generation capabilities in LocalAI and I've noticed a potential limitation in the current Web UI regarding the ACE-STEP-TURBO model.
The ACE-STEP-TURBO model supports a "cover mode", which inherently requires an "Audio Reference" to function correctly. However, looking at the audio generation section in the UI, the available fields are:
It appears there is no option or input field to upload or provide a path to an audio file to act as the "Audio Reference".
My questions are:
Thank you for your hard work on this amazing project!
All reactions