1First, hear the two product targets
The local-listening version is the clean, polished sound: low noise, strong body, and a more intimate pulse. It is the version you judge as an audio product.
The online-playback version is built for streaming platforms. It may preserve more wideband heartbeat context because the platform encoder needs that context. It is not simply louder or cleaner; it has a different job.
2Peak normalization: why the tallest point is not enough
Peak means the tallest single sample in the waveform. Peak normalization finds that tallest point and scales the whole file so that point reaches the target.
That sounds reasonable until one unusually tall heartbeat decides the gain for the entire recording. Weak heartbeats stay weak relative to that tall one, because the whole file moved together.
A limiter after normalization is also not a miracle. It can stop the tallest peaks, but if you push into it too hard, strong heartbeats become flat and quiet residue rises with the makeup gain.
The better unit is one heartbeat event. If a real heartbeat is weak, it can be raised. If a tiny tail is merely attached to the previous beat, it should not be promoted into a fake main beat.
One peak scales everything; weak beats stay weak.
Each beat is solved by role and headroom.
3Why heartbeat-event loudness is better
A heartbeat event is one meaningful heartbeat action: its beginning, hit, body, and tail. It can contain S1, S2, or a split detail, but the question is always the same: what role does this sound play in the rhythm?
If one full heartbeat is weak, it may deserve gain. If a small tail after the previous heartbeat is weak, it should usually remain a tail. File-level tools cannot tell the difference.
Event-level loudness feels more natural because it does not reward every quiet thing. It asks whether the quiet thing is actually a heartbeat that should become stronger.
4Pipeline, marker, and unit: the terms behind event-level loudness
Pipeline means the ordered processing workflow: detect the heartbeat, find its footprint, classify the event, shape it, control loudness, then export the product.
A marker is a time label. For example: onset begins here, attack peaks here, body ends here. A marker does not make sound by itself; it tells the processor where a sound region is.
An enriched event is a detected heartbeat event after we add timing, role, peak time, and S1/S2/split information. A heartbeat unit is the group used for loudness decisions, so a main beat and a close dependent detail do not get treated as unrelated sounds.
In short: markers are labels, enriched events are labeled heartbeat actions, and units prevent the loudness tool from boosting the wrong fragment.
5Closed-form event gain
Closed-form means the algorithm calculates the gain directly instead of trying values again and again.
For each eligible heartbeat, we know the current waveform and the extra contribution we want to add. The algorithm asks: what is the largest gain that keeps every sample below the target peak?
That gives a one-pass answer for that event. It is more predictable than recursion, and it avoids using a limiter as the main volume tool.
6Why close attacks and split tails need protection
A close attack is a second hit or detail that happens very close to the previous heartbeat. A split tail is a real small S2 or tail detail, but it belongs to the previous heartbeat's shape.
If the loudness system mistakes that dependent detail for a full new heartbeat, it can boost it too much. The listener then hears a strange electric or synthetic sound, because a tail has been turned into a fake main beat.
The safer workflow groups close hits together. Inside that group, only the dominant heartbeat gets the main gain. Dependent details are protected instead of inflated.
7Clean enhanced loudness is different
The clean enhanced product is the pure, polished version. It can use adaptive enhancement and a soft limiter because it is trying to sound beautiful as a file, not necessarily survive the harshest streaming transcode.
But final fullband normalization can lift tiny high-frequency residue around every heartbeat. The cleaner strategy is to strengthen the real heartbeat event first, catch peaks gently, and then reduce high-frequency residue.
The principle is consistent: make the real heartbeat stronger on purpose, not every quiet thing in the file.