Everything gets streamed these days, and almost all streaming platforms use lossy data encoding: mp3, AAC, OGG etc (OMG!)
And these lossy codecs do a pretty remarkable result of preserving the quality, considering they strip out anywhere between 60-90% of the original data. But it’s not perfect - far from it ! In this show we talk about:
- How your audio can be damaged by lossy data encoding
- Why this happens
- How you can test for it yourself
- Why it’s worth doing
- How it can help you hear the artefacts for yourself
- Whether we should optimise our audio to avoid them and (most importantly!)
- The two simple steps you can take to minimise these problems and help your audio SURVIVE streaming
Links
Video - Is Billie Eilish too loud ?
TMS #36 - Data versus dynamic compression - and sponges
Spotify recommends -2 dB True Peak for loud uploads
TMS #25 – Stereo, mid-side and how to process them
Really interesting show thank you both.
I will have to mull this over a lot and decided how this will affect my production decisions. It makes me wonder if the algorithms are miss reading “deliberate” distortion and adding compensation artefacts which actually then equate to more distortion.
Reminds me of the distortion we used to get from early rock CDs where the bass speakers sounded like they were going to rip apart.
The deep learning audio mangler has to become real!
I think the same tech we have for images can be applied - we'd just have to convert a lot of audio into spectrogram images, and train the neural network on that library, then feed it a spectrogram of whatever we want mangled, and it should give us another spectogram, that we can resynthesize into audio.
Somebody make this a thing!