Add crates/denoise with DeepFilterNet dependency #3246
+713
−30
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Create new denoise crate that wraps DeepFilterNet libDF for audio denoising - Uses git dependency on DeepFilterNet v0.5.6 tag - Provides clean interface with Denoiser struct for processing audio frames - Supports 48kHz sample rate with 480 sample hop size - Includes process_frame() for single frame and process() for batch processing - All tests passing Co-Authored-By: yujonglee <yujonglee.dev@gmail.com>
Contributor
Author
🤖 Devin AI EngineerI'll be helping with this pull request! Here's what you should know: ✅ I will automatically:
Note: I can only respond to comments from users who have write access to this repository. ⚙️ Control Options:
|
✅ Deploy Preview for hyprnote-storybook canceled.
|
✅ Deploy Preview for howto-fix-macos-audio-selection canceled.
|
✅ Deploy Preview for hyprnote canceled.
|
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
![graphite-app[bot]](https://avatars.githubusercontent.com/in/158384?s=60&v=4){kind=small}
- Return error when input length is not a multiple of hop_size - Add test for invalid length error case - Add benchmark suite following aec crate pattern - Benchmarks: initialization, process_frame, process (10/100/1000 frames), throughput Co-Authored-By: yujonglee <yujonglee.dev@gmail.com>
- Load 48kHz audio from english_1 test data - Benchmark with actual audio samples instead of zero-filled buffers - Results: ~284µs per frame, ~3.45 Msamples/s throughput Co-Authored-By: yujonglee <yujonglee.dev@gmail.com>
- Add default-model-ll feature flag for DeepFilterNet3 low-latency model - Update cfg attributes to support both default-model and default-model-ll - Benchmark comparison shows default model is faster for batch processing: - Default: 3.45 Msamples/s, ~50s for 1h audio - Low-latency: 1.84 Msamples/s, ~94s for 1h audio - Low-latency model has reduced algorithmic latency for real-time use Co-Authored-By: yujonglee <yujonglee.dev@gmail.com>
Co-Authored-By: yujonglee <yujonglee.dev@gmail.com>
Contributor
Author
|
Devin is archived and cannot be woken up. Please unarchive Devin if you want to continue using it. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Adds a new
crates/denoisecrate that wraps DeepFilterNet libDF for audio noise suppression. The crate provides a cleanDenoiserstruct interface for processing audio frames at 48kHz with a 480-sample hop size.Key implementation details:
process_frame()for single-frame processing andprocess()for batch processingUpdates since last revision
default-model-llfeature flag for low-latency model (reduced algorithmic latency for real-time use)process()now returnsInvalidFrameSizeerror instead of silently truncatinghypr-datacrateBenchmark results (48kHz audio):
default-modeldefault-model-llNote: "Low-latency" refers to reduced algorithmic lookahead, not faster processing. Use
default-modelfor batch processing.Review & Testing Checklist for Human
process()now returns error for non-multiple of hop_size inputs instead of silent truncationSuggested test plan:
cargo test -p denoiseto verify unit tests passcargo bench -p denoiseto verify benchmark runsNotes
This crate is not yet integrated into the audio pipeline - it's just the foundation crate. Integration with listener/audio processing would be a follow-up.
Requested by @yujonglee
Link to Devin run: https://app.devin.ai/sessions/6128c661afaa4d9ca4bbf3b90071483a