Lossy compression that discards harmonic detail changes timbre, not just loudness

What it is

Speech or music run through low-bitrate lossy compression — an older voice codec, a heavily compressed audio format, or a degraded codec mode used over Bluetooth — often ends up sounding "thin," "boxy," or slightly metallic. Those descriptions point to a change in timbre, not a drop in volume. Compression artifacts primarily attack harmonic structure first, not overall loudness.

This is easy to misjudge: turning down device volume and compression loss both make sound feel "less full," but compression changes the character of the sound — turning the volume back up cannot recover harmonic detail that was already discarded.

Why it happens

Mainstream perceptual audio codecs rely on a psychoacoustic masking model: the algorithm predicts which frequency regions can have their energy discarded or coarsely quantized without being noticed, because they would be masked by stronger, co-occurring frequency content anyway — concentrating the bitrate budget on what's actually audible. The lower the bitrate, the less harmonic detail this model can afford to keep, typically starting with high-order harmonics and mid-level harmonics judged perceptually insignificant.

The problem is that harmonic structure is exactly the spectral fingerprint that determines timbre (see related entry): once high-order harmonics are stripped, the source's original spectral envelope gets truncated and narrowed, and the audible result is a change in timbre — the sound loses its original fullness and material quality, rather than simply "getting quieter." Loudness perception is driven mainly by total signal energy, which compression generally leaves largely unchanged, so loudness stays roughly the same while the distribution of energy across the spectrum is what actually shifts.

Studying it

A common approach compresses the same source material at different bitrates or with different codecs, then has listeners rate subjective quality (e.g., mean opinion score) or perform an ABX discrimination task (judging which of A or B matches reference X) to measure whether the ear can reliably tell the versions apart; spectral analysis comparing harmonic amplitude distributions before and after compression is used to link the drop in subjective rating to the objective amount of harmonic loss.

Common independent variables: encoding bitrate, codec type, how harmonically rich the source material is. Common dependent variables: subjective quality rating, ABX discrimination accuracy, magnitude of spectral-envelope difference before vs. after compression.

Where it stops holding

• At the higher bitrates typical of mainstream codec settings today, harmonic loss is usually small enough that most listeners cannot reliably detect it in blind tests; this mechanism shows up most clearly at low bitrates — older or simplified voice-call codecs, audio aggressively compressed to save bandwidth, or a Bluetooth codec mode downgraded due to bandwidth constraints.
• Sounds with richer harmonic structure that depend more on high-frequency detail (strings, cymbals and similar percussion) tend to show more noticeable timbre change under compression than harmonically simple sounds; a test result from one type of material shouldn't be generalized to all material.
• This entry is about compression's effect on timbre; if a compression pipeline also introduces noticeable loudness normalization or dynamic range compression, that is a separate loudness-level change that should be assessed on its own rather than conflated with the timbral artifact.

Applying it

• When choosing a codec and bitrate for audio assets that serve an identification function — brand sound marks, notification sounds meant to be sorted into categories at a glance — don't judge by file size or overall volume alone; specifically check whether the harmonic structure survives compression well enough to preserve the original recognizability.
• If a product's audio must pass through a transmission path known to degrade it significantly (low-bandwidth voice calls, certain Bluetooth codec modes), re-test whether key alert sounds are still correctly classified after that specific path, rather than signing off only on the uncompressed design files.
• Verification: process the finished alert sounds through the actual codec parameters the target transmission path will use, and have listeners judge under blind conditions whether the compressed version still maps to the same event category as the uncompressed one — use classification accuracy, not "sounds fine," as the acceptance criterion.

Related

• Same group: A3.12.1 Timbre is set by the harmonic structure above the fundamental, and is the primary cue for telling sources apart · A3.12.3 Realistic sound effects imply physical material by matching its harmonic signature · A3.12.4 Voice quality shapes listeners' subjective judgment of content credibility
• Nearby: A3.04 Auditory masking · A3.09 Environmental noise and signal-to-noise ratio
• Search terms: lossy audio compression · perceptual audio coding · mean opinion score · codec artifacts