Struct SpecAugment 

pub struct SpecAugment { /* private fields */ }

SpecAugment: Data augmentation for speech recognition (Park et al., 2019).

Applies time warping, frequency masking, and time masking to spectrograms.

Methods

• Time Warping: Warps spectrogram along time axis
• Frequency Masking: Masks F consecutive frequency channels
• Time Masking: Masks T consecutive time steps

Reference

• Park, D., et al. (2019). SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition. Interspeech.

Implementations

impl SpecAugment

pub fn new() -> Self

Create SpecAugment with default parameters.

Default: 2 frequency masks (F=27), 2 time masks (T=100)

pub fn with_params( num_freq_masks: usize, freq_mask_param: usize, num_time_masks: usize, time_mask_param: usize, ) -> Self

Create with custom parameters.

pub fn with_mask_value(self, value: f32) -> Self

Set the mask value.

pub fn apply( &self, spec: &[f32], freq_bins: usize, time_steps: usize, ) -> Vec<f32>

Apply SpecAugment to a spectrogram.

Arguments

• spec - Spectrogram as flat vector [freq_bins * time_steps]
• freq_bins - Number of frequency bins
• time_steps - Number of time steps

Returns

Augmented spectrogram.

pub fn freq_mask( &self, spec: &[f32], freq_bins: usize, time_steps: usize, ) -> Vec<f32>

Apply only frequency masking.

pub fn time_mask( &self, spec: &[f32], freq_bins: usize, time_steps: usize, ) -> Vec<f32>

Apply only time masking.

pub fn num_freq_masks(&self) -> usize

pub fn num_time_masks(&self) -> usize

Trait Implementations

impl Clone for SpecAugment

fn clone(&self) -> SpecAugment

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SpecAugment

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for SpecAugment

fn default() -> Self

Returns the “default value” for a type.