Why Does My Track Sound Quieter on Streaming Services?
The short answer
Most streaming platforms practice loudness normalization: they measure each track's overall loudness and adjust its playback gain so that different tracks reach the listener at a similar loudness. A master that is louder than the platform's reference gets turned down at playback. The file itself is untouched; only the volume applied during playback changes.
The measurement underneath
Normalization is possible because loudness measurement was standardized. ITU-R BS.1770 defines the algorithm — K-weighted, gated, integrated over the program — that produces a single LUFS value per track, plus the true-peak measurement that bounds how high the waveform actually swings. Because every compliant meter computes the same number, a platform can measure millions of uploads consistently and adjust each toward a common reference. The mechanics of the measurement are covered in LUFS vs. dBFS.
Each platform sets its own reference level and publishes it in its upload or delivery documentation. The targets differ between services and have changed over time, so the current number belongs to the platform's own docs, not to a third-party summary.
What this changed in practice
Before normalization, a louder master was simply louder on playback, which rewarded pushing masters toward maximum loudness at the cost of dynamic range — the pattern often called the loudness war. Under normalization the reward disappears: a heavily limited master and a dynamic master of the same song are turned to the same effective loudness, and the limited one keeps its reduced dynamics without any volume advantage. Professional discussion and technical documents around loudness practice are maintained in venues like the Audio Engineering Society, which publishes standards and technical documents in this area.
What normalization does not do
- It does not remaster audio. Gain is applied uniformly; dynamics, EQ, and clipping already baked into the file remain.
- It does not always turn tracks up. Platforms reliably turn down material louder than the reference; whether quiet material is turned up varies by service and by user settings.
- It does not equalize sections within a track. The integrated measurement yields one value for the whole program; quiet intros stay quiet relative to loud choruses.
- It is not always on. Some platforms allow listeners to disable it, and some playback contexts skip it.
Practical takeaways for anyone uploading audio
Meter your work in LUFS with a BS.1770-compliant meter, check the current documentation of each platform you deliver to, and leave true-peak headroom as those documents specify — lossy encoding can push intersample peaks higher than the original file shows. Beyond that, loudness is a mixing and mastering decision, not a compliance trick: normalization has mostly removed the incentive to chase level for its own sake. For the units involved from first principles, start at what is a decibel.