Automated Gabor-transform identification succeeded on 67 bird species but struggled with low-frequency-dominated songs

Evidence

Automating identification based on Gabor transform has been recently proposed by Connor et al. (2012). These authors have utilized the Gabor frequency-amplitude spectrum (GFAS) and Gabor time-amplitude profile (GTAP). These two parameters have been combined and compared using a distance metric with a library of known birdsongs/calls. The identification of 67 bird species on 568 digital recordings of songs/calls was successful, although songs or calls of species dominated by low frequencies were more difficult to identify than species characterized by higher frequencies. — Almo Farina, Soundscape Ecology, 2054

Categorization

• QUESTION: Why do low-frequency-dominated calls resist the Gabor approach — a genuine biological/acoustic gap worth chasing for the methods framework.
• XREF: connects to: Automated vocalization identification still fails under masking from geophony and anthrophony — Almo Farina, Soundscape Ecology: Principles,…, 2049; Song Scope frog identification failed on roughly half of recordings — Almo Farina, Soundscape Ecology: Principles,…, 2055; Polyphonic concurrent sources make automated music structure recognition difficult (+1 more)

Broader Connections

• Sound spectrum analysis depends on banks of filters to decompose signals #expands (0.78)
• Polyphonic concurrent sources make automated music structure recognition difficult #corroborates (0.75)