Stereo width in AI music: what we measured, and when widening helps

What the measurement showed

For 20 Suno WAV downloads the side-to-mid energy ratio averaged −9.3 dB, which is the normal range for commercial pop and rock masters. Stereo correlation was positive and healthy on all 20. In plain terms: the WAVs were not narrow.

The 'narrow' reputation most likely comes from the MP3 exports and from earlier model versions. Lossy encoders at low bitrates use joint-stereo tricks that collapse the sides. If you are working from a WAV, treat width as fine until a meter says otherwise.

When widening actually helps

• A track that measures narrower than about −14 dB side/mid in the mid and high bands, with the low end already mono.
• A sparse arrangement where the reverb tails are all centred.
• A mono source that was generated or exported as dual-mono (both channels identical). Widening a true mono file only adds artefacts; it needs a mix, not a master.

When it hurts

Widening below about 120 Hz makes the bass unstable and cancels in mono. Widening a vocal-centred track pulls the vocal back. And any widener that works by delaying or inverting one channel breaks mono compatibility: on a phone speaker, a Bluetooth speaker, a club system summed to mono, the widened parts thin out or disappear.

Our engine widens per band and only when the measured width is below the reference range for the style. On the 20 test tracks it mostly left the stereo image alone, which is the correct decision when nothing is wrong.

Width is the one thing AI generators already get right. Fix the tone first.

How to check mono compatibility

Sum the master to mono and listen for elements that drop in level or change tone. A correlation meter reading below zero for more than a moment means something will cancel. Most streaming players do not sum to mono, but most phones do.