- Migrated to pyproject.toml
- Extracted noisereduce and visrecognize to optional dependencies
- Support for python 3.11
- Added close_seconds_filter option to BaseConfig and decorators on recognize and align functions
- Fingerprint, fine_align, single-threaded bug
- VisualRecognizer multiprocessing on macs
- normalize option in BaseConfig, Turns off all normalization if False
- Updated dependencies
- dockerfiles to run tests
- unprocessed argument to write_shifts_from_results and write_shifted_file
- Order or arguments for write_shifts_from_results
- Multichannel output option from aligners
- Updated dependencies
- API to match new API from noisereduce
- Removed python 3.7 support, added 3.10
- using effects module from pydub for normalize
- Bug in making all output audio files the same length
- Find find_matches_not_in_file_shifts but
- setup.py file
Changed
- Removed Audalign class, changed to module methods
- Recognitions are done with recognizer objects
- Configuration is done through config objects held by recognizers
- Spectrogram freq_threshold works on hertz rather than specgram bands
- Fingerprint alignments are not independent anymore, so removed load_fingerprints option in alignments
- load_fingerprints option in alignments
- Better handling of second match in rankings
- number of fingerprints print in alignments
- run_align.py script. CLI with argparse to run alignment easily
- runtime divide by zero warning
- butter filter error with 0 threshold
- target aligning prelim checks for files outside against dir
- Alignment cases with no matches
- alignments are all positive now. You can easily align the files in a DAW by placing the files at the given time mark
- removed list around fingerprints recognition locality_frames_setting
- sped up finding audio files
- pretty printers for recognitions and alignments in audalign object
- recalc shifts from alignment results
- write_shifts_from_results. For use with recalc shifts or using different source files
- sped up remove noise by about 80%
- doesn't try to write m4a's
- rank alignment added to all recognitions and alignments
- rank alignment function
- noise remove write extention
- noise remove destination directory support
- can't read ext in writes, so unsupported file types are written to wavs
- raises error if destination directory doesn't exist at the start of alignments
- noise remove is prettier
- correcognize and correcognize spectrogram filter_matches defaults to 0
- updated docstrings
- inverted overlap ratio calculator
- Fixed bug with no results in correcognizes and visrecognize
- Correcognize offsets
- Correcognize_spectrogram offsets
- Added multiprocessing to correcognize
- Added multiprocessing to correcognize_spectrogram
- Sped up correlation max finding
- Correlation_Spectrogram: correlation technique, but with the spectrogram
- Better docs for align functions
- Plotter for correlations
- Recognize max_lags locality bug
- Scaling factor in correlation normalized for length and bit depth
- Default locality_filter_prop lowered to include more results
- correlation locality
- Multiprocessing for visual recognition/alignment on Linux works
- Multiprocessing for visual recognition/alignment optimized better for Windows
- Lots more neat tests
- multiprocessing works for Windows. Linux forces single threaded.
- By my tests, Windows multiprocessing and Linux were the same speed.
- fine_align: to give a precise alignment after a rough one
- Lets you specify match_index for which rough alignment match to use
- Lets you specify width of alignment for fine alignment
- Lets you specify which alignment technique to use (visual not implemented yet)
- max_lags for recognize and correcognize
- visrecognize to reduce pickling between processes
- pytest specifies tests dir; number of processes bumped to 10
- align_files: like align, but takes two or more filenames/paths to align
- align_files tests
- addopts to pytest.ini file
- tests: all write to tmpdir except for saving fingerprints
- Pydub linux file reading bug: throws IndexError where it should be CouldntDecodeError
- Locality_filter_prop: filters locality tuples by proportion of highest confidence to tuple confidence within each offset
- Locality tuples include individual confidences at the end of each tuple
- Locality, improperly sorted through tuple noise
- Locality, wasn't creating correct tuples
- metadata function
- Correlation based alignment and correcognize
- frequency threshold getter and static setter
- "use_fingerprints" to "technique" that takes a string
- changed filter fingerprints in align to none to tell if user input is supplied
- Bumped urllib and Pillow
- Messed up version nums
- fingerprinting and alignment windows
- Sped up fingerprinting in tests
- Documentation all same style
- write shifted files loophole
- read write destination
- Github Actions
- Locality Fingerprinting
- recognize bug
- tests
- Args and kwargs for noise reduce
- Visual Alignment weighting
- Try except for write in aligns
- Writing total track
- target align bug
- Visual Calc_mse
- Visual Image Resizing
- Visual Volume Floor
- Tons of Stuff
- Many Tests
- There are previous versions, but I don't think anybody would care about them