Lost in Compression: Who’s Heard and Who’s Blurred in Digital Voice Communication?

Digital communication has become essential post-COVID-19, yet speech transmission technologies introduce unintended distortions that may disproportionately affect certain voices. This study investigates whether three widely used compression and decompression algorithms (codecs)—OPUS, Adaptive Multi-Rate (AMR), and CODEC2—introduce structural bias by differentially degrading male and female voices. Using a dataset of 2,953 speakers, we applied 14 audio quality metrics to assess differences in voice degradation across various bitrate settings. Our results reveal consistent, metric-dependent disparities by sex, with OPUS showing a shift in bias at mid-range bitrates, and AMR and CODEC2 exhibiting persistent but variable bias across all settings. Results indicate that female voices experience greater degradation across multiple measures, potentially impacting listener perception and communication clarity. These findings highlight the importance of codec design in equitable voice representation and call for greater attention to inclusion in speech processing systems.