WaveWash
An open-source audio data cleaning and deduplication tool for non-destructive preprocessing before ML training. It includes four tuning presets for ASR, TTS, speaker recognition and music generation, supports DC offset repair, clipping repair, EBU R128 loudness normalization, SNR estimation and duplicate sample detection, and provides a zero-configuration local Web UI and audit logs.
Timeline (last 90 days)
WaveWash is an open-source audio data cleaning and deduplication tool used to automatically perform audio quality control and data preparation before ML training.
WaveWash adopts a non-destructive preprocessing design, in which all processing is done on copies, and original files are never modified.
WaveWash includes built-in tuned preset profiles for four scenarios: ASR (Whisper), TTS, speaker recognition, and music generation.
WaveWash can automatically correct DC offset, repair mild clipping, and normalize loudness according to the EBU R128 standard.
WaveWash sorts passing audio into the clean/ directory, classifies rejected files by reason into _error/ subdirectories (such as low_snr/, clipping/), places duplicate samples into the _duplicates/ directory, and deletes no files.
WaveWash provides audit log summaries in three formats: CSV, JSON, and Markdown.
WaveWash's Web UI is packaged directly inside the Python package, requiring no additional Node.js build step.
WaveWash can be installed via 'pip install wavewash', and the Web UI is launched with 'wavewash app', with the default address being http://127.0.0.1:8420.
WaveWash supports a command-line run mode, with the command 'wavewash run --input ./raw_audio --output ./clean_audio --purpose asr'.
WaveWash is currently in an early alpha stage, with version number v0.1.0a1.
1 more timeline events
All Facts (10)
WaveWash is an open-source audio data cleaning and deduplication tool used to automatically perform audio quality control and data preparation before ML training.
50%UnverifiedWaveWash adopts a non-destructive preprocessing design, in which all processing is done on copies, and original files are never modified.
50%UnverifiedWaveWash includes built-in tuned preset profiles for four scenarios: ASR (Whisper), TTS, speaker recognition, and music generation.
50%UnverifiedWaveWash can automatically correct DC offset, repair mild clipping, and normalize loudness according to the EBU R128 standard.
50%UnverifiedWaveWash sorts passing audio into the clean/ directory, classifies rejected files by reason into _error/ subdirectories (such as low_snr/, clipping/), places duplicate samples into the _duplicates/ directory, and deletes no files.
50%UnverifiedWaveWash provides audit log summaries in three formats: CSV, JSON, and Markdown.
50%UnverifiedWaveWash's Web UI is packaged directly inside the Python package, requiring no additional Node.js build step.
50%UnverifiedWaveWash can be installed via 'pip install wavewash', and the Web UI is launched with 'wavewash app', with the default address being http://127.0.0.1:8420.
50%UnverifiedWaveWash supports a command-line run mode, with the command 'wavewash run --input ./raw_audio --output ./clean_audio --purpose asr'.
50%UnverifiedWaveWash's GitHub repository address is https://github.com/BoobalanArjunan/wavewash, and its PyPI page is https://pypi.org/project/wavewash/.
50%