Audio

Audio tracks and clips

A video file with sound arrives as a linked pair: picture on a video track and sound on the audio track under it, moving and trimming together. Music and voice dropped in on their own are audio clips like any other.

Mute

On the track header. Audio tracks only — a video track's clips are picture-only, so there is nothing on one for a mute to act on.

Lock

Stops the track being edited while you work above it.

Waveform

Drawn from the file's own peaks, at a fixed resolution, so zooming the timeline redraws from the same data rather than decoding again.

Levels

The base item on an audio clip's rack: volume and pan, both keyframable. There is no separate inspector tab for sound because this is it.

The sound rack

Same rack as the video side — rows in processing order, an eye on each that keyframes, parameters you scrub or type — running a Web Audio graph instead of a shader chain.

Parametric EQ

A notch or a boost at a frequency you choose, with a width.

Bass & Treble

A tilt. Two shelves, one at each end.

High-Low Pass

A cliff. Everything above or below the cutoff goes.

Reverb

A room. The one that keeps sounding between words.

Delay

Repeats, at an interval, decaying.

Flange & Chorus

A modulated delay — thickening, or a sweep.

Stereo Mixer

Balance, width, and the channels swapped or summed.

Backwards

Plays the clip's sound in reverse.

Why it behaves differently from the video rack

A shader chain is rebuilt from scratch every frame, because the frame is thrown away afterwards and nothing is cheaper than not remembering. An audio graph is the opposite: a delay line is its accumulated past and a reverb is its tail, so rebuilding one per frame would not be slow so much as wrong — a tail that restarts sixty times a second.

• Parameters are pushed rather than reapplied. Only what moved is written, and it is ramped.
• Bypassing a row is a crossfade between the processed path and the dry one, not a teardown. Otherwise switching an effect off part-way through a clip would click, and drop whatever was decaying.
• The graph is only rebuilt when the rack's shape changes — a row added, removed or reordered. That is also the one edit you expect to hear as a change rather than as a glide.

Auditioning an effect

In the palette, an audio tile draws the clip's own waveform at rest and plays that passage on a loop through the real graph when you point at it, with the waveform giving way to the spectrum of what is coming out. A tone control is a shape you can see: a high-pass is a cliff, a parametric EQ is a notch, bass and treble is a tilt.

Reverse

A clip running at a negative rate plays backwards, and the sound goes with it. A media element has no reverse gear, so a reversed clip stops being driven by its element and starts being driven from a decoded buffer, played from the far end.

That costs a full decode of the file, which is why it is cached and shared: two reversed clips of one file are one buffer, and trimming, splitting or duplicating does not pay again. Speed and ramps are on Timeline.

Video effects driven by sound

Audio Spectrum and Audio Waveform draw what a track is playing. They ask which audio track to listen to, and what they hear has been through that clip's whole rack — every effect, every keyframe on them, its level — so a spectrum of audio you have filtered shows the filtering.

What the export does with sound

The finished soundtrack is rendered offline in one pass rather than recorded in real time — so it cannot glitch on a busy tab, and the parts a player cannot do at all (a reversed shot, an 8× shot) are ordinary. The same graph builder that made the reverb you heard makes the reverb in the file. See Export.