Struct SpectralDesc

pub struct SpectralDesc { /* private fields */ }

General object holding all the spectral descriptor.

Holds 3 spectral descriptors together. It would be better conceptually to have 3 different spectral descriptor objects, but this avoids re-computing the same FFT three times.

Current spectral descriptors are spectral centroid, spectral rolloff and spectral flatness (see values_object for a further description of the object.

All descriptors are currently summarized by their mean only.

Implementations

impl SpectralDesc

pub const WINDOW_SIZE: usize = 512usize

pub const HOP_SIZE: usize = 128usize

pub fn get_centroid(&mut self) -> Vec<Feature>

Compute score related to the spectral centroid values, obtained after repeatedly calling do_ on all of the song’s chunks.

Spectral centroid is used to determine the “brightness” of a sound, i.e. how much high frequency there is in an audio signal.

It of course depends of the instrument used: a piano-only track that makes use of high frequencies will still score less than a song using a lot of percussive sound, because the piano frequency range is lower.

The value range is between 0 and sample_rate / 2.

pub fn get_rolloff(&mut self) -> Vec<Feature>

Compute score related to the spectral roll-off values, obtained after repeatedly calling do_ on all of the song’s chunks.

Spectral roll-off is the bin frequency number below which a certain percentage of the spectral energy is found, here, 95%.

It can be used to distinguish voiced speech (low roll-off) and unvoiced speech (high roll-off). It is also a good indication of the energy repartition of a song.

The value range is between 0 and sample_rate / 2

pub fn get_flatness(&mut self) -> Vec<Feature>

Compute score related to the spectral flatness values, obtained after repeatedly calling do_ on all of the song’s chunks.

Spectral flatness is the ratio between the geometric mean of the spectrum and its arithmetic mean.

It is used to distinguish between tone-like and noise-like signals. Tone-like audio is f.ex. a piano key, something that has one or more specific frequencies, while (white) noise has an equal distribution of intensity among all frequencies.

The value range is between 0 and 1, since the geometric mean is always less than the arithmetic mean.

pub fn new(sample_rate: u32) -> AnalysisResult<Self>

Errors

This function will return an error if there is an error loading the aubio objects

pub fn do_(&mut self, chunk: &[f32]) -> AnalysisResult<()>

Compute all the descriptors’ value for the given chunk.

After using this on all the song’s chunks, you can call get_centroid, get_flatness and get_rolloff to get the respective descriptors’ values.