Speech intelligibility alone cannot capture loss of speaker identity in codecs

Evidence

The STI measure considers only the intelligibility of speech. This has not been found to be an adequate analysis system for modern telecommunication systems, where the speech codecs may, in the worst case, for example, make the speaker unidentifiable, although the intelligibility of speech is good. To estimate other factors in transmitted speech, more advanced objective analysis methods have been developed. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2437

Categorization

• XREF: Connects to modern telephony/VoIP codec analysis and the broader limitation of single-metric audio quality assessment. — connects to: Male speaker identity survives when the voice fundamental is filtered out; Speech intelligibility and naturalness are distinct perceptual dimensions that do not track each other — David M. Howard, Jamie Angus, Acoustics and Psychoacoustics, 2398
• SEED: Possible essay angle: the gap between intelligibility and identity in speech codecs, and what acoustics reveals about the limits of optimization metrics.
• corroborated: 2 independent passages support this claim — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2437

Broader Connections

• Speech intelligibility and naturalness are distinct perceptual dimensions that do not track each other #complicates (0.85) — David M. Howard, Jamie Angus, Acoustics and Psychoacoustics, 2398