Hi-fi internet radio, explained

Why radio sounds flat in the first place

Two separate processes damage a radio stream before it reaches your speakers:

1. Broadcast processing. Stations compress hard so the signal stays loud and consistent on cheap receivers and in cars. Quiet passages get pushed up and loud ones squashed, so the music ends up with less dynamic range than the recording had.
2. Lossy coding. MP3, AAC and Opus throw away high-frequency information the encoder believes your ears will not notice. At 128 kbps that is a lot of information; at 320 kbps it is far less.

Most players simply decode the result and send it to your output. A hi-fi player rebuilds what it can.

The chain, stage by stage

Audio expander

The expander is the opposite of a compressor: below its threshold, it reduces the level of quiet material further. Applied to heavily compressed radio, that increases the distance between quiet and loud passages again, so the music breathes instead of sitting at one constant level. It runs in an AudioWorklet, so the processing happens in real time on the audio thread rather than in JavaScript.

Harmonic exciter and air shelf

Lossy codecs remove high frequencies; an exciter generates harmonics that were not transmitted but are musically consistent with the fundamentals that were. Combined with a high shelf, this makes lossy highs feel restored. It is synthesis, not recovery — but the perceptual result is a stream that no longer sounds like it is hiding behind a blanket.

AutoEQ for headphones and earbuds

Headphones all add their own frequency-response colour. The player ships 23 correction profiles (Sennheiser HD 600/650, Beyerdynamic DT 770/990, Audio-Technica M50x, Bose QC35, Sony WH-1000XM4 and WF-1000XM4, AirPods Pro, Galaxy Buds, Grado, AKG, HIFIMAN and more) applied to a 10-band EQ, and it can import any AutoEQ ParametricEQ, GraphicEQ or CSV file you point it at. Imported profiles stay in your browser.

AGC and the limiter

Auto volume normalisation (AGC) works on the raw pre-FX signal so it is independent of the enhancer, and it targets a consistent average loudness instead of chasing peaks. The final limiter is a lookahead design with a −1 dBFS ceiling: when the chain adds gain this catches the peaks transparently, and the LIM indicator lights when it is working.

Metering that shows what changed

Four meters and an impact readout turn “it sounds better” into something measurable: spectrum, LED-style VU, K-12 dynamics (average level plus crest factor), stereo image (M/S and correlation), frequency response of both the raw and processed signals, and a summary of how much dynamics and loudness the chain added or removed in dB.

What no processor can do

Anyone selling an audio tool should be clear about this:

• Deleted frequencies cannot be recovered. If the encoder removed 16 kHz, no expander will bring it back. The exciter adds harmonics that are consistent with what remains, which sounds much better than nothing but is not the same thing.
• A bad stream stays a bad stream. 128 kbps with heavy broadcast compression is a hard starting point. The highest-quality stream a station offers is always the better choice.
• Processing is not a fix for bad masters. A well-mastered track that was compressed again for broadcast is damaged by that processing, not by the codec alone — and the expander is aimed exactly at that damage.

Practical settings that work

• Leave Enhance ON and start with the presets for a station type, then adjust.
• Use AutoEQ if you listen on headphones or earbuds with a known profile.
• Keep AGC on for mixed playlists; turn it off on a well-mastered station that already holds its level.
• Watch the impact readout: if it reports that dynamics barely changed, the stream was not very compressed to begin with, and a lighter touch will sound more natural.

If your main question is whether radio can replace a subscription, read the honest comparison. If you listen in a shop or café, start with music licensing for venues.