Verdict: Unlikely.
Theoretical Background
The term "lossy" refers to loss of information - that is, the decompressed audio does not exactly match the uncompressed raw audio - and as a result, may contain artifacts not present in the original. However, this data loss does not translate into missing audio that needs to be "filled-in" - the audio is just different. So at least the explanation for this claim is not valid.
Even listening to CD-quality music, there is noise and artifacts introduced by the speakers, surrounding environment, and internal sources (eg, imperfect hearing). Our brain naturally processes audio input all the time to address this, a phenomenon known generally as top-down processing. Common examples in the auditory domain include audio continuity, phonemic restoration, and duplex perception. For this claim to be valid, it should be demonstrated that the sort of artifacts introduced by lossy compression result in a need for mental audio processing sufficiently beyond the background rate to cause detectable fatigue.
Listener fatigue can certainly result from noticeable degradation of audio input, such as that caused by high levels of noise, hearing loss, and range compression. However, according to the current state of the art Framework for Understanding Effortful Listening (FUEL) and related Ease of Language Understanding (ELU) model, additional processing is not required when audio is intelligible as-is. In unnecessarily technical jargon:
If the RAMBPHO-delivered sub-lexical information matches a
corresponding syllabic phonological representation in semantic LTM,
then lexical access will be successful and there is no need for
top-down processing.
Thus, fatigue is not anticipated for intelligible quality audio. The FUEL framework also incorporates motivation:
... fatigue may hinge on motivation, and the control and management of
goals insofar as expending effort can be fatiguing if goals are
externally imposed but not when activities are self-initiated and
meaningful.
In other words, listener fatigue may be a factor when actively listening to low-quality audio, such as trying to understand song lyrics in a noisy environment. Conversely, passively listening to music is unlikely to cause fatigue (presumably even if audio quality is poor).
The ELU model posits that audio processing in the brain has a certain capacity - referred to as working memory. Accordingly, even actively listening to low-quality audio should not result in fatigue as long as the audio processing demand does not exceed this capacity.
Empirical Evidence
Fatigue can be measured as performance decline over time, or less reliably, as a subjective self-report. However, a necessary (but not sufficient) prerequisite of fatigue, is the relative expenditure of effort, the lack thereof indicating no cause for fatigue. Mental effort can be measured in a variety of ways, such as increased response time, dilated pupils, brain waves on EEG, or blood flow on fMRI.
Another way of measuring effort, the dual-task paradigm, asks subjects to perform a second task while actively processing audio (usually speech). For example, performance on a secondary visual processing task suffers more when parsing low-quality than high-quality audio (Pals, Sarampalis, & Başkent, 2013), demonstrating that more effort is expended. However, as audio quality increases, performance stops improving, suggesting no additional effort is required at audio quality levels well below what the human ear can distinguish - in other words, even noticeably lower quality audio (above 8 channels in the below image) can be processed with no additional mental effort relative to the raw (labeled Unpr below) audio.
Similar dual-task studies show no additional mental effort or fatigue when hearing-aid digital noise reduction is turned off (Hornsby, 2013), from moderate levels of reverberation (Picou et al, 2019), typical levels of classroom background noise (McGarrigle, 2017), and moderate range compression (Stone et al, 2009), presumably as long as working memory capacity is not exceeded.
These results are intuitive not just because of the legions of current MP3 music-listeners, but also several previous generations of listeners of poor quality music from radio, record players, audio tapes, and other formats far more lossy than typical MP3 audio. Additional mental effort, as per the dual-task paradigm, would affect people's ability to listen to music and simultaneously drive, walk, work, or study, which is often the opposite of what we see in real-life.
Notes:
- Measures of effort other than dual-task may produce different results, reflecting differences in underlying constructs. For example, pupillometry provides evidence of fatigue at lower levels of noise than dual-task (eg, Brännström et al, 2021; McGarrigle et al, 2017), as it may be a more sensitive measure. There is too much literature to review here.
- The studies reviewed here have many limitations, such as the use of relatively short audio segments, the focus on speech rather than music, degrading audio quality in different ways than lossy compression would, measuring effort rather than fatigue, etc.
- Researchers and engineers concerned with MP3 compression quality are rarely also interested in cognitive psychology, so direct tests of this claim are hard to come by. An example that comes close is a small study by Antons et al (2012) (experiment III), that used a lossy compression codec called G.722.2, in an active listening task using single-word audio clips, and was able to detect brain activity using EEG at quality levels not noticeably lower than uncompressed audio. The authors suggest that this activity may represent effort that can lead to fatigue, though this study did not test for fatigue:
... we could show that the pattern of brain activation related to
processing degradations consciously can also be detected in trials
which are not reported as degraded on a subjective level. ... we
assume that cognitive load and fatigue might increase when being
exposed to small degradations for a long period of time ...
