Voice Telephony for Individuals with Hearing Loss: The Effects of Audio Bandwidth, Bit Rate and Packet Loss
Linda Kozma-Spytek, Paula Tucker, Christian Vogler · 2019 · Proceedings of the 21st International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2019) · doi:10.1145/3308561.3353796
Summary
This paper presents three experiments conducted over three years with a total of 114 individuals with hearing loss (68 cochlear implant users and 46 hearing aid users) and 12 hearing controls, investigating how audio quality parameters in telecommunications networks affect the accessibility of voice phone calls. Telephone communication has long been problematic for people with hearing loss — many report it as frustrating, tiring, and embarrassing. The core problem is that traditional telephone networks (PSTN) transmit only narrowband (NB) audio in the 300-3,400 Hz range, cutting off approximately 20% of speech information at higher frequencies that are critical for intelligibility. Wideband (WB) audio codecs, also called HD Voice, extend this range to 50 Hz-7 kHz and have been deployed on 4G/VoLTE networks, but cross-carrier interoperability remains poor and calls between different carriers typically fall back to narrowband. The three experiments systematically tested: (1) narrowband versus wideband audio at maximum codec quality with 42 cochlear implant users; (2) the effect of varying bit rates across six codec conditions (including the low rates typical of mobile networks) with 36 CI and HA users; and (3) the combined effects of bandwidth and packet loss (0%, 3%, 20%) with 36 people with hearing loss and 12 hearing controls. AT&T Labs carried out all signal processing to ensure stimuli mirrored real-world network conditions.
Key findings
Across all three experiments, wideband audio consistently and significantly improved outcomes for people with hearing loss. In Experiment 1, WB audio improved speech recognition by 10.4 percentage points (75.7% vs 65.2% words correct, p<0.0001) and reduced perceived mental effort by 16.3 points on the SMEQ scale (p<0.0001). Experiment 2 replicated these findings with real phones and self-selected listening levels, and revealed that bit rate matters significantly: the lowest narrowband mobile codec (AMR-NB at 5.90 kbps) produced both worse speech recognition and lower quality ratings than standard PSTN quality (G.711 at 64 kbps), meaning mobile phones are objectively harder for people with hearing loss than landlines. Experiment 3 showed that packet loss devastates accessibility: at 20% packet loss with narrowband audio, speech recognition for people with hearing loss dropped to 42.9% (vs 77.6% at 0% loss), while hearing participants dropped from 98.1% to 83.0%. Wideband audio maintained its advantage even under packet loss, particularly on managed networks with 3% loss. Critically, hearing participants were also significantly affected by bandwidth and packet loss, but much less so — their speech recognition remained above 83% even in worst-case conditions, while people with hearing loss fell to 43-51%. The interaction between bandwidth and packet loss was significant for quality ratings in the hearing loss group but not in the hearing group.
Relevance
This research provides the first systematic, empirical evidence that telecommunications audio quality parameters — which are entirely under the control of carriers and standards bodies — have a direct, measurable impact on accessibility for the 30+ million Americans with hearing loss. The practical implications are clear and actionable: telecoms should favour wideband audio over narrowband (a win-win for all users), ensure cross-carrier WB interoperability (which has languished since a 2014 FCC petition), use bit rates higher than 5.90 kbps for narrowband mobile audio, and deploy managed networks with quality-of-service guarantees to minimize packet loss. The finding that the typical mobile codec (AMR-NB at 5.90 kbps) performs worse than a traditional landline call is particularly alarming given the rapid displacement of landlines. For policy, the paper supports making wideband audio a requirement rather than a recommendation in standards like Section 508 and EN 301 549. For accessibility practitioners, this work demonstrates that infrastructure-level audio quality is an overlooked but significant accessibility barrier — simply making speech louder is not enough when the audio bandwidth strips out the frequencies people with hearing loss need most.
Tags: hearing loss · telecommunications · cochlear implant · hearing aid · audio quality · VoIP · wideband audio · speech intelligibility · accessibility standards
Standards referenced: Section 508 · EN 301 549