Media::WaveformSummary Struct Reference

Detailed Description

Per-channel minimum and maximum of the samples in fixed buckets of frames.

Drawing a waveform zoomed out far enough that one pixel column spans thousands of samples otherwise costs a full pass over the file per redraw, whatever the width of the window: the work is proportional to the length of the sound, not to the number of columns. Reducing whole buckets instead divides that by the bucket size.

Indexed in source frames – the same coordinates AudioFile::ViewHandle takes – so loops, start offsets, tempo and the device pixel ratio never enter into it: a range has passed through all of those before it gets here.

Built once, when the file has finished decoding, and never written again. It is passed around as a shared_ptr<const WaveformSummary> and read from the waveform threads without synchronisation of its own; an earlier attempt at this grew its table while readers were walking it, and that is what made it unreliable.

Public Member Functions
const FloatPair *	channel (int c) const noexcept

Static Public Member Functions
static int64_t	bucketFor (int64_t frames, int32_t channels) noexcept

Public Attributes
int64_t	bucket {}
	Frames per bucket. A power of two, at least 64.
int64_t	bucketCount {}
	Buckets per channel. The last one may cover fewer than bucket frames.
int32_t	channels {}
int64_t	frames {}
	Frames the summary accounts for.
std::unique_ptr< FloatPair[]>	data
	channels * bucketCount pairs, channel-major.

Member Function Documentation

bucketFor()

static int64_t Media::WaveformSummary::bucketFor ( int64_t  frames, int32_t  channels  )

inlinestaticnoexcept

Bucket size keeping the table within a fixed budget whatever the length of the sound. Short sounds get the finest bucket, which is what decides the zoom at which the summary starts paying: it only helps once a pixel column spans several buckets, so 64 frames means it engages from roughly 256 samples per pixel – well inside the range one actually works at. Long sounds trade that back for memory: an hour of stereo lands on 128 frames and 20 MB, ten hours on 1024 and 25 MB.

---

The documentation for this struct was generated from the following file:

• WaveformSummary.hpp