Fixed-width filter bands mismatch hearing’s finer low-frequency resolution

Evidence

A typical resolution in complex QMF processing is K = 64 frequency bands, which, with a sampling rate of 44.1 kHz, results in bandwidths of approximately 345 Hz. As discussed earlier, the frequency resolution of hearing functions follows ERB or Bark bands (see Section 9.4.3 on page 167). Thus, this resolution is insufficient for the lowest frequencies. Therefore, the implementations discussed in Breebaart et al. (2005, 2007) and Herre et al. (2012) have applied a cascaded filter bank at the lowest frequency bands to obtain a higher frequency selectivity, which shown in Figure 15.12. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2138

Categorization

• XREF: Connects filter bank design to ERB/Bark auditory frequency-resolution models from Section 9.4.3 of this same book.
• QUESTION: Cascaded filter banks at low bands trade off latency and complexity — worth understanding the cost/benefit tradeoff in spatial audio codecs.