Octavelabeled

Octavelabeled is a labeling scheme used in audio signal processing and music information retrieval that assigns spectral content to octave bands rather than to exact frequencies. The term describes a representation in which each time frame of an audio signal is characterized by the presence or energy of octave-spaced bands, typically arranged from lower to higher octaves. The approach emphasizes scale invariance to pitch shifts and can facilitate comparisons across musical transpositions or different tuning systems.

In practice, octavelabeled features are generated by computing a time-frequency representation (such as a short-time Fourier transform) and grouping the resulting frequency bins into octave bands. Each frame yields a vector of octave-band energies or activation probabilities, often with optional normalization or smoothing. The number of octaves and the choice of a reference frequency determine the labeling granularity. The resulting representation can be used as input to machine learning models for tasks like genre classification, instrument recognition, onset detection, or audio indexing.

Octavelabeled representations relate to other spectral features by providing a coarse, perceptually motivated alternative to fine-grained frequency analyses. Compared with chroma features, octavelabeled data preserve octave information beyond pitch class, while offering robustness to small detunings. Limitations include reduced resolution for closely spaced harmonic content and sensitivity to the chosen octave boundaries or sampling rate.

Octavelabeled is not a universal standard but a descriptive term for this kind of octave-based spectral labeling used in specialized audio analysis workflows.