---

name: py-audio-analysis description: "Use pyAudioAnalysis for audio feature extraction, classical audio classification/regression, segmentation, diarization, legacy CLI commands, and audio I/O troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

py-audio-analysis

Use this repo skill when a task involves pyAudioAnalysis, classical audio/speech/music analysis, short- or mid-term acoustic features, sklearn-backed audio classifiers/regressors, segmentation, silence removal, speaker diarization, audio thumbnails, or pyAudioAnalysis command-line tasks.

First checks

1. Confirm the package is installed. Public package name: pyAudioAnalysis; import package: pyAudioAnalysis.
2. For a safe installation and dependency check, run:

   python scripts/check_pyaudioanalysis_env.py --pretty
   

   Run it from this skill directory or give the script path explicitly.
3. If a task asks whether this skill matches a particular checkout or package version, read references/repo-provenance.md.
4. For the package/module map, dependency summary, and workflow overview, read references/package-overview.md.
5. For cross-cutting install/import, dependency, plotting, and legacy CLI failures, read references/troubleshooting.md.

Minimal import/API smoke for user environments:

from pyAudioAnalysis import ShortTermFeatures, MidTermFeatures, audioBasicIO

Do not use python -m pyAudioAnalysis.audioAnalysis as the default CLI path: version 0.3.14 has legacy top-level imports in audioAnalysis.py. Use the cli-and-io sub-skill for safe installed-package CLI inspection or prefer Python APIs when possible.

Route by task

Feature extraction, spectrograms, chromagrams, beats

Read sub-skills/feature-extraction/SKILL.md when the user asks to:

• read audio and convert stereo to mono before analysis;
• compute ShortTermFeatures.feature_extraction(...) rows/names;
• compute MidTermFeatures.mid_feature_extraction(...) or directory feature matrices;
• generate spectrogram/chromagram arrays without GUI side effects;
• write feature matrices to NPY/CSV or debug unexpected feature shapes.

Classification and regression

Read sub-skills/classification-regression/SKILL.md when the user asks to:

• train a pyAudioAnalysis classifier from class-folder WAV data;
• classify one file or a folder with svm, svm_rbf, knn, randomforest, gradientboosting, or extratrees;
• train or apply audio regression models;
• understand model artifact files, MEANS sidecars, pickle safety, or sklearn compatibility;
• build a bounded synthetic smoke test for a classifier workflow.

Segmentation, diarization, silence removal, thumbnails

Read sub-skills/segmentation-diarization/SKILL.md when the user asks to:

• train/use HMM segmentation from WAV + segment annotations;
• run mid-term classifier-based segmentation over an audio file;
• remove silence and interpret kept time spans;
• run speaker diarization or evaluate diarization purity;
• generate music thumbnails;
• split Audacity-style labels into per-label audio clips.

Legacy CLI, audio formats, conversion, visualization, dependency probes

Read sub-skills/cli-and-io/SKILL.md when the user asks to:

• construct or inspect audioAnalysis.py commands and flags;
• troubleshoot ModuleNotFoundError: ShortTermFeatures from legacy CLI imports;
• check WAV/AIFF/MP3/AU/OGG support, pydub, eyeD3, ffmpeg, or avconv;
• convert media to WAV or reason about conversion side effects;
• run feature-visualization commands safely in a headless environment.

Important operating constraints

• Required backend is CPU. The selected pyAudioAnalysis workflows are NumPy/SciPy/scikit-learn/hmmlearn workflows; no GPU/accelerator backend is required.
• MP3/media conversion is a system-dependency concern. WAV workflows can pass while MP3 conversion still needs ffmpeg or avconv.
• Many legacy CLI commands plot with Matplotlib or write files beside the input/model path. Prefer API calls with plot=False and explicit temporary output directories when automating.
• Training and model-loading APIs use pickle artifacts. Load only trusted model files.
• Time windows are usually seconds in high-level APIs/CLI arguments, but lower-level feature functions accept sample counts. The feature sub-skill calls this out explicitly.
• Several maintainer shell commands assume external datasets. Treat them as evidence for command shape, not as default runtime checks.

Bundled helper scripts

• scripts/check_pyaudioanalysis_env.py: shared import/dependency/legacy-CLI/feature smoke check.
• sub-skills/feature-extraction/scripts/feature_smoke.py: synthetic or WAV-based feature matrix smoke.
• sub-skills/classification-regression/scripts/classification_smoke.py: bounded synthetic classifier train/classify smoke.
• sub-skills/segmentation-diarization/scripts/segmentation_smoke.py: safe silence-removal and optional HMM/diarization smoke.
• sub-skills/cli-and-io/scripts/inspect_cli.py: installed-package legacy CLI inspector.
• sub-skills/cli-and-io/scripts/audio_io_smoke.py: audio I/O and optional media-tool probe.

Avoid this skill when

• The task is deep neural audio training/inference (for example Whisper, NeMo, CLAP, PyTorch TTS, or diffusion audio); use a model-family-specific skill instead.
• The task is only generic NumPy/SciPy signal processing without pyAudioAnalysis APIs, model artifacts, data formats, or CLI behavior.
• The user is asking to edit this repository's source code, package metadata, or tests rather than use pyAudioAnalysis as a library; route to a Python repository-maintenance skill.