In this lesson
Why delivery specs save the song
Before anyone hears your WAV, it is encoded, decoded and turned up or down, and a delivery spec is the list of what it has to survive.
Updated · 6 min read · Lesson 44 of 44 in Mixing & Mastering
In short
- After upload, your master is measured, encoded to a lossy format, decoded on a device and turned up or down by normalization.
- Encoding and reconstruction can both push peaks above the WAV, so the true-peak ceiling has to leave room for them.
- Deliver what each destination asks for, meter an encoded copy before release, and keep an unlimited premaster for everything else.
In this lesson (8)
The upload that sounded different
You finish the master. It sounds clear and deep in the studio, and you send a 24-bit WAV to your distributor. Weeks later, on a streaming app, the top end is harsher and the loudest kicks crackle. You reload the WAV, check the meters and find no anywhere.
The listener never heard your WAV. On the way, it was measured, encoded to a lossy format, decoded on a phone and turned up or down by . A delivery spec is the list of things your master has to survive on the way.
Why it matters
Your file is processed after you let go
Services do their own processing. Spotify says it prepares the compressed and lossless streams itself from one high-quality stereo master per track, and its advice is written for lossy formats such as Ogg Vorbis and AAC. Apple publishes the commands it uses to encode masters to 256 kbps AAC at ingest, so you can audition that encode yourself (Apple, 2021).
Each of those steps can change your peaks. EBU Tech 3343 lists lossy coding among the processes that create above the original sample level, along with the converters and converters that follow.
The science
What each step can do to the master
Measurement. The service reads integrated loudness with ITU-R BS.1770 and sets a playback gain. On Spotify's Normal setting, a master louder than -14 is turned down. A quieter one is raised only until its peaks reach -1 dB, so a dynamic master with high peaks may not be raised at all. The Loud setting is the exception: it raises soft tracks regardless of true peak and catches them with a .
Encoding. A lossy encoder stores an approximation of your waveform that should sound the same. The decoded wave is a different wave, and it can peak higher than the original, more so on loud, dense masters. That is why Apple tells you to check the encoded file for clipping, as well as the 24-bit master.
Decoding and conversion. The device rebuilds the continuous wave from samples. That wave can crest between samples, and any stage with a fixed maximum level clips what goes past full scale.
That is why the true-peak ceiling is lower than 0. The numbers come from the documents that set them:
| Item | Spec |
|---|---|
| File | One high-quality stereo master per track (Spotify) |
| Resolution | The original 24-bit PCM file (Apple Digital Masters) |
| True peak | Below -1 dBTP, or -2 dBTP if louder than -14 LUFS (Spotify) |
| Ceiling | -1 dBTP for PCM, -2 dBTP for some broadcast codecs (EBU) |
| Clipping | None, in the WAV or in an encoded test (Apple) |
Different destinations have different limits. Vinyl struggles with loud high frequencies and with bass that is out of between the channels, so cutting engineers usually want a separate, less limited master. Ask before you send one file everywhere.
Try it
Audition the encode
0 of 6 steps done
On loud material the encoded copy often reads higher than the WAV. The difference signal shows you where the codec works hardest, usually in the top end and around .
Common mistake
One master for every destination
One loud master goes everywhere: to streaming, to the vinyl cutter and into your archive. Limiting cannot be undone, so keep an unlimited premaster for remasters, vinyl and formats you have not thought of yet.
The other mistake is trusting normalization to fix level problems. It changes gain and nothing else. It will not limit a loud master, it will not clean up overs that are already in the file, and on the Normal setting it will not raise a quiet master past the point where its peaks would clip.
Takeaway
Deliver to the destination
Set a true-peak ceiling of -1 , or -2 dBTP for masters louder than -14 LUFS going to Spotify. Before release, meter an encoded copy as well as the WAV. Deliver each destination the file it asks for, and keep a premaster the final limiter has not touched. How true peak works in detail is in the lesson on true peak after encoding.
Clipping
Cutting off the top of a waveform at a fixed level. Hard clipping flattens the peaks; soft clipping rounds them, which adds fewer harsh harmonics.
Loudness normalization
A streaming service turning each track up or down so everything plays back at a similar loudness. A louder master is simply turned down further.
True peak
The highest level a waveform reaches between samples once it is converted to analog or decoded from a lossy file. It can sit above the highest sample value, so meters estimate it by oversampling.
Inter-sample peak
A peak in the reconstructed waveform that falls between two samples, so the highest sample value under-reads it. True-peak meters find it by oversampling.
Sample rate
How many times per second a digital system measures the waveform. 48 kHz means 48,000 measurements a second.
LUFS
Loudness Units relative to Full Scale. A loudness measurement weighted toward how the ear hears. Streaming services use integrated LUFS, an average over the whole track, to match playback level.
Limiter
A very fast compressor with a very high ratio. It stops peaks from going over a set ceiling.
Phase
Where a wave is in its cycle at a given moment. Two copies of a sound that are out of phase partly cancel where one rises while the other falls.
Transient
The short, sharp start of a sound, like the crack of a snare or a pick hitting a string. It carries a lot of how hard a sound feels and how quickly the ear recognizes it.
dBFS
Decibels relative to full scale, the largest value a fixed-point file or converter can hold, so levels below it are negative numbers. Inside a floating-point mix a signal can read above 0 dBFS and stay intact; it clips where it leaves float or passes through a plugin that clips. dBTP is the same scale measured as true peak, and it can read above 0.
3 questions
Check yourself
Sources
Show all 6Hide sources
- Apple. (2021). Apple Digital Masters [Technology brief]. www.apple.com/apple-music/apple-digital-masters/docs/apple-digital-masters.pdf
- European Broadcasting Union. (2023). EBU R 128: Loudness normalisation and permitted maximum level of audio signals. EBU. tech.ebu.ch/publications/r128/
- European Broadcasting Union. (2023). Tech 3343: Guidelines for production of programmes in accordance with EBU R 128. EBU. tech.ebu.ch/docs/tech/tech3343.pdf
- International Telecommunication Union. (2023). Recommendation ITU-R BS.1770-5: Algorithms to measure audio programme loudness and true-peak audio level. ITU. www.itu.int/rec/R-REC-BS.1770/
- Katz, B. (2015). Mastering Audio: The Art and the Science (3rd ed.). Focal Press.
- Spotify for Artists. Loudness normalization on Spotify. support.spotify.com/artists/article/loudness-normalization/
Lesson 44 of 44
You reached the end of Mixing & Mastering
That was the last of 44 lessons in this path. The next path is Audio Science, 14 lessons.
Audio Science, lesson 1Sampling is taking photos of airDigital audio is not a staircase. The sample rate sets the highest frequency a file can hold, and the converter plays one smooth wave through the samples.Back to the Mixing & Mastering pathAll pathsAll lessons
Previous: Sequencing changes perceived quality

Written by Virzy Guns
Songwriter and producer, founder of Virzy Guns Production. Now building HealingWave.