Logarithmic quantization suits speech but distorts music beyond tolerance

Evidence

The high levels of a signal are thus represented with larger quantization steps than the lower levels. Logarithmic quantization generates unacceptable distortion in a music signal. However, the artefacts are not prominent with speech signals, and logarithmic quantization, in practice, gives a dynamic range of about 12 bits. Thus, when each sample is quantized using 8 bits, and sampled at the rate of 8 kHz, the resulting rate of transmission is 64 kbit/s, which is acceptable in wired communication, but relatively high for wireless telephony. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2217

Categorization

• XREF: Ties to the earlier chapters on the matching of coding (mu-law companding) to signal statistics — a classic signal-class-properties point.
• SEED: The difference in how speech vs music tolerates quantization distortion is a neat angle for a piece on why codec design is media-specific.