Struct nnnoiseless::DenoiseFeatures

pub struct DenoiseFeatures {
    pub x: [Complex32; 481],
    pub p: [Complex32; 481],
    pub ex: [f32; 22],
    pub ep: [f32; 22],
    pub exp: [f32; 22],
    /* private fields */
}

Contains the necessary state to compute the features of audio input and synthesize the output.

This is quite a large struct and should probably be kept behind some kind of pointer.

Fields

x: [Complex32; 481]

The Fourier transform of the most recent frame of input.

p: [Complex32; 481]

The Fourier transform of a pitch-period-shifted window of input.

ex: [f32; 22]

The band energies of x (the signal).

ep: [f32; 22]

The band energies of p (the signal, lagged by one pitch period).

exp: [f32; 22]

The band correlations between x (the signal) and p (the pitch-period-lagged signal).

Implementations

impl DenoiseFeatures

pub fn new() -> DenoiseFeatures

Creates a new, empty, DenoiseFeatures.

pub fn features(&self) -> &[f32]

Returns the computed features.

pub fn shift_input(&mut self, input: &[f32])

Shifts our input buffer and adds the new input to it. This is mainly used when generating training data: when running the noise reduction we use DenoiseFeatures::shift_and_filter_input instead.

pub fn shift_and_filter_input(&mut self, input: &[f32])

Shifts our input buffer and adds the new input to it, while running the input through a high-pass filter.

pub fn compute_frame_features(&mut self) -> bool

Computes the features of the current frame.

The return value is true if the input was pretty much silent.

pub fn pitch_filter(&mut self, gain: &[f32; 22])

Applies a filter to the audio, attenuating pitches that have poor correlation with the pitch-lagged signal.

Trait Implementations

impl Clone for DenoiseFeatures

fn clone(&self) -> DenoiseFeatures

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.