Hearing sensitivity to frequency can be emulated by filtering and weighting short-time power spectra

Evidence

The computation of an auditory spectrum begins with the power spectrum computation. The input signal is first divided into, say, 25-ms-long time frames that are then multiplied by a suitable window function, like a Hamming window. Thereafter, a short-term power spectrum is computed for each time frame using the DFT, which is implemented using the fast Fourier transform (FFT). This processing does not reflect the frequency-dependent sensitivity of hearing. In practice, the sensitivity can be emulated at any stage of the computation, but conceptually, it should be emulated at the beginning by filtering the input signal. The Matlab script yields a coarse approximation of the inverse of the equal loudness contour at 60 dB SPL (Figure 9.2) by high-pass filtering the signal at the beginning and then multiplying the short-term power spectra by frequency-dependent weights. As a consequence, both the poor sensitivity of hearing to low frequencies and the increased sensitivity around 4 kHz, resulting from the ear canal resonance, are emulated roughly. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 1586

Categorization

• XREF: Connects to Chapter 9’s equal loudness contours and Chapter 10’s excitation patterns from earlier in this book; also to psychoacoustic modeling in audio engineering. — connects to: Auditory spectrum computation begins with FFT then emulates hearing sensitivity via filtering and weighting. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 1586