Have you ever wondered how loud your audio productions should be? So have we, because it's a quite tricky question to answer!
On the one hand, you want to make sure your productions are loud enough. For example, they should still appear at a reasonable level in noisy environments (car, airplane, etc.), and their loudness should be comparable to that of other programs, no matter whether those are produced by yourself or by others.
On the other hand, you certainly don't want to compromise sound quality for sheer loudness. We have discussed in Audio loudness measurement and normalization with EBU R128 how the Loudness War of recent decades has resulted in a loss of dynamics (and thereby of sound quality) in professional audio. We have also discussed a solution to this problem: the loudness unit (LU), a measure that allows us to predictably normalize sound to an actually perceived loudness rather than to peak levels that bear no resemblance to perception.
But what's a suitable loudness value for your own productions? "How many LU" should your podcast or video have? In this blog post, we'll try to answer this question. Specifically, we'll look at the requirements of audio that is distributed online and consumed on mobile players.
The -23 LUFS recommendation in EBU R128
The loudness unit (LU, or LUFS with reference to Full Scale) was originally proposed in the EBU R128 recommendation. This recommendation is quite clear on what constitutes a suitable loudness value: it proposes -23 LUFS as the new center of the audio levelling universe
(p.4).
As usual, however, things are not quite as simple as that. The EBU R128 recommendation was primarily concerned with television, but radio programs or podcasts are often consumed on mobile devices and therefore in noisier environments. Also, we are still in a transition period: although loudness normalization will without doubt become the standard practice over the next few years, the majority of audio programs do not yet follow EBU R128 or similar recommendations.
After all, most of the hardware players expect very loud audio and don't provide enough gain for programs normalized to -23 LUFS. Hopefully this will change in near future.
Which loudness value do producers choose now?
At Auphonic, we have analysed the loudness of some existing podcasts and radio programs, in order to get a better understanding of typical loudness values currently used for audio distributed online. We performed our analysis by means of the loudness command line tool that is included in the libebur128 library (r128gain would be another suitable tool for this purpose). The selection of audio programs is the same as for the podcast comparison series. Here are the results:
| Podcast | Channels | Program loudness | Loudness range (LRA) | True peak |
|---|---|---|---|---|
| CRE 188 | 2 | -18.8 LUFS | 6.6 LU | -0.2 dBTP |
| techzing 162 | 2 | -15.7 LUFS | 3.4 LU | 0.4 dBTP |
| Hoaxilla 70 | 2 | -18.9 LUFS | 6.9 LU | 1.1 dBTP |
| Planet Money Podcast | 1 | -20.2 LUFS | 7.2 LU | 1.1 dBTP |
| Startups For the Rest of Us 61 | 2 | -15.3 LUFS | 3.4 LU | -0.8 dBTP |
| Alternativlos 21 | 2 | -17.0 LUFS | 2.6 LU | 0.2 dBTP |
| Bits und so 272 | 1 | -19.3 LUFS | 3.6 LU | 0.3 dBTP |
| Elementarfragen 9 | 2 | -17.5 LUFS | 3.0 LU | -0.3 dBTP |
| Ubuntu UK Podcast (high) | 2 | -19.0 LUFS | 10.4 LU | 3.0 dBTP |
| Ubuntu UK Podcast (low) | 1 | -22.0 LUFS | 10.2 LU | 2.2 dBTP |
| Fanboys Podcast 49 (MP3) | 1 | -19.4 LUFS | 3.5 LU | 0.6 dBTP |
| Fanboys Podcast 49 (HE AAC) | 2 | -16.1 LUFS | 3.5 LU | 1.4 dBTP |
| This American Life 420 | 1 | -18.7 LUFS | 6.4 LU | 3.0 dBTP |
| ARD Radio Tatort | 2 | -22.6 LUFS | 12.5 LU | -0.2 dBTP |
| Deutschland Radio Wissen | 1 | -15.0 LUFS | 3.4 LU | 2.0 dBTP |
| Freakonomics | 2 | -16.5 LUFS | 5.1 LU | 0.8 dBTP |
| OE1 Geschichte Podcast | 1 | -26.5 LUFS | 5.9 LU | -3.6 dBTP |
| Wait Wait ... don't tell me | 1 | -23.8 LUFS | 10.7 LU | -1.4 dBTP |
| Average values | -19.0 LUFS | 6.0 LU | 0.5 dBTP |
Note that the average program loudness is -19 LUFS, 4 LUFS louder than the EBU recommends. Also, the average true peak level of the analyzed podcasts lies at +0.5dB. In other words, the average podcast clips. This illustrates the problem of the Loudness War, and the need for additional dynamic headroom, quite nicely.
For the sake of completeness, the loudness range (LRA) of the selected programs, a measure we have discussed in an earlier blog post, is also included in the above table.
What about software?
A popular tool among podcasters is The Levelator, which allows to balance loudness differences within an audio program, similar to Auphonic's Adaptive Leveler. We have analyzed a number of files produced with the Levelator and found that they typically end up around -18 LUFS.
It's interesting to note that an attempt to ensure ReplayGain and EBU R128 compatibility also resulted in a suggestion of -18 LUFS as a suitable compromise. More information on this can be found in a Hydrogenaudio forum thread and in an AES paper.
The Levelator software tool.
It's also interesting to see how Apple handles the problem of loudness normalization. The Sound Check feature in iTunes analyzes the loudness of newly added songs and stores this information in an ID3 tag or in the iTunes Music Library database. Your iPod then uses this information to compensate loudness differences between different tracks during playback. According to Thomas Lund, a researcher at tc.electronic, the resulting loudness corresponds to around -16 LUFS.
Is -23 LUFS loud enough for podcasts and mobile audio?
EBU R128 was primarily devised for television, but podcasts are often consumed on lower-quality audio systems and in noisier environments. So how does the -23 LUFS value perform under these circumstances?
At Auphonic, we have conducted some informal trials to answer this question. We have simply done what many people do: listen to podcasts on laptops, cellphone loudspeakers and headphones while simultaneously conducting everyday activities such as cooking or using public transportation. To us it seemed that a loudness of -23 LUFS is not sufficient under these circumstances. Most noticeably, speech intelligibility clearly suffers at that level and many hardware players don't provide enough gain to make the audio loud enough.
Thomas Lund also concludes that -23 LUFS "is too low for pod/mobile platforms". He proposes the following distribution levels, which correspond well to our own experience:
- For podcasts: -16 to -23 LUFS
- For mobile audio: about -16 LUFS
See his great article Audio for Mobile TV, iPad and iPod for detailed information and practical tests regarding loudness on mobile devices.
Furthermore a new Music Loudness Alliance group tries to establish a loudness metadata system similar to Sound Check or ReplayGain which is enabled by default. Then all audio which is too loud, or without metadata, will be punished automatically.
What about radio?
it's important to distinguish between analog FM radio and digital radio.
There is not much hope that FM radio will switch to loudness normalization any time soon, partly because of concerns that lower signal levels might result in less signal reach, which in turn could translate to fewer listeners and less money from advertising.
However, digital radio is supposed to switch to loudness normalization in future and people from the EBU PLOUD group are working on that! For example, Norway already applies loudness normalization to radio, where stations use a target value of -15 LUFS (read DAB Norway - implementation of loudness normalization for details).
Conclusion
Although the EBU R128 recommendation proposes -23 LUFS as a meaningful loudness value for audio productions, this target value does not yet perform well under all circumstances, such as the noisy listening environments typical for mobile audio. It's still very early days for loudness normalization, and it will probably take some time until definite standards emerge to account for the plurality of listening circumstances. At Auphonic we are monitoring these developments very closely, in order to provide you with best-practice audio processing options, which give your productions a professional sound quality and make them compliant with the latest industry standards.
For now, you can choose from the following loudness target values in the Auphonic production form:
Loudness target values in the Auphonic production form.
- For audio consumed on mobile devices, we recommend -16 LUFS, which is also the default value in the production form. If you are unsure as to which value to use, we suggest you try this one first.
- To be similar to the ReplayGain reference level, use -18 LUFS.
- If you want to stay compliant with EBU R128, choose -23 LUFS. To follow the related ATSC A/85 specification, go for -24 LUFS. The subtle differences between the two are described in an earlier blog post.
- In addition, we offer several other target values (-13, -15, -20, -31 LUFS) to fully cover the whole spectrum of reasonable loudness targets.
UPDATE 2015:
The Audio Engineering Society (AES) published an official podcast/streaming loudness recommendation:
Recommendation for Loudness of Audio Streaming and Network File Playback
Summary of the recommendation:
- Integrated Loudness should be between -16LUFS and -20LUFS
- the Maximum True Peak Level should be <= -1dBTP
UPDATE 2017:
Various audio-related companies did add loudness recommendations recently:
- Apple Music, Google: -16 LUFS
- Amazon Alexa, Spotify and Tidal: -14 LUFS
- Youtube: -14 LUFS
UPDATE 2018:
Apple also mentions a loudness and true peak target in their
Podcasts Authoring Best Practices:
- Overall Loudness should remain around -16 LUFS, with a +/- 1 LU tolerance
- True Peak Values should not exceed -1 dBTP
UPDATE 2019:
For further information, please also see our new article
The New Loudness Target War
.