Audio streaming and on-demand playback are a significant part of most people’s daily sources of entertainment. Whether commuting, in the shower, doing chores, or just relaxing in general, we are always listening. One factor that seems to constantly annoy listeners and detract from this popular form of entertainment is inconsistent loudness levels.
When we talk about controlling and measuring loudness in a piece of audio, sound engineers often measure it in LUFS or Loudness Units relative to Full Scale. This is measuring loudness against the maximum amount an audio system can handle. LUFS is always shown as a negative number, and the closer the negative number LUFS reading is to the number zero, the higher the average loudness of that audio is.
In recent technical recommendations published by the Audio Engineering Society, AES for short, it’s clear that the approach to measuring and managing audio loudness needs updating. Technical Document 1008, found here, was published in September 2021. In this document, the AES contends that speech-centric audio content should be normalized to -18 LUFS, whereas music should more likely be normalized to -16 or -14 LUFS to allow for a wider dynamic range. Before TD1008, recommendations were more general and focused on staying within a range of -20 to -16 LUFS.
To reach the desired loudness level, audio is normalized downward if its loudness is above the target level or upward if it is below. The loudness of most music we listen to is high, so a downward ‘gain’ is added. Some music may require a different adjustment (which would be an upward normalization). This would be audio content with a great peak-to-loudness ratio. Here, peak limiting would be recommended, or if the dynamic audio quality is adversely affected, only some normalization would be added.
Before being distributed to the public, music needs to be normalized. This is for an optimal listener experience and the artist’s artistic intent. Album normalization is often hard to achieve on the radio where songs from different artists are played. The tracks are adjusted separately or “track normalized” in this circumstance. Each track is raised or lowered by different quantities to achieve a similar loudness. However, this likely will alter an artist’s intent by making some songs sound louder (or weaker) than they were intended to be by the artist.
An audience should receive as consistent an audio experience as possible, regardless of the platform they're listening on or the form of content. When it comes to content creation, creators should want to focus on the loudness levels that they’re setting. Whether mixing a podcast or an album, the listening experience can be rather jarring if loudness levels fluctuate noticeably.
Services use a BS 1770 loudness meter to measure loudness. They start by measuring integrated loudness, and if that loudness is too close to the full measure of loudness, then engineers need to compress or limit the loudness levels. This may come at the cost of dynamic range and overall clarity. The ratio between the highest peak level and the integrated loudness is named the peak to loudness ratio.
Streaming services often have varying loudness normalization from each other. Users can simply turn the volume up or push it into a limiter if a platform has content generally at a loudness level of -18LUFS. The only problem with this is that issues of peak to loudness ratio will be affected, and the listening experience can be uncomfortable.
To summarize, -16 to -18 LUFS appears to be the optimal level for streaming platforms to use for podcasting and music. This loudness range covers most music forms and the speaking in podcasts (as well as advertising snippets). Some users may make their own adjustments for their own personal experience, but these levels make it comfortable to listen to on most audio systems. We strongly suggest that engineers interested in improving their audio workflows refer to the AES TD1008 recommendations for loudness levels used in internet streaming and on-demand distribution.