Skip to content

feat(streaming): add backend-agnostic live audio editing - #311

Draft
ryanontheinside wants to merge 2 commits into
mainfrom
feat/audio-edit-backend-agnostic
Draft

feat(streaming): add backend-agnostic live audio editing#311
ryanontheinside wants to merge 2 commits into
mainfrom
feat/audio-edit-backend-agnostic

Conversation

@ryanontheinside

Copy link
Copy Markdown
Collaborator

Summary

  • add backend-agnostic live waveform edit state across ACE-Step and SA3 streaming requests
  • preserve source samples outside selected edit regions at the final streamed-audio boundary
  • treat an enabled empty region set as an all-preserve mask and immediately restore the source buffer
  • validate only explicit extension tails and preserve ordinary backend canvas padding
  • retain the hidden StreamDenoise audio-edit node registration and cover left/right extension paths

Correctness details

The request snapshot travels through StreamDenoise and SlotRequest into each backend's mask or inpainting path. Final output is constrained by the intersection of that emerged request snapshot and the current edit selection after runner crossfades. Edit activation is serialized with slice publication so stale unmasked slices cannot overwrite the restored client mirror.

Verification

  • Direct StreamDenoise import and node id check
  • CPU unit suites: 34 passed across live audio edit, wire contract, and client SDK
  • git diff --check

GPU/model-level inference was not available locally; preservation, mask polarity, snapshot propagation, ordinary interior edits, empty masks, and left/right extension behavior are covered at the CPU/session delivery boundary.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant