Struct DcHighPassFilter

Source

pub struct DcHighPassFilter { /* private fields */ }

Expand description

DC offset removal and high-pass filtering with streaming state.

Implements the DC offset removal specification:

Removes DC offset using exponential moving average (EMA)
Applies cascaded Butterworth high-pass filtering (defaults to 4th order @ 80 Hz)
Maintains filter state across chunks for streaming continuity
Achieves <2ms latency target per 500ms chunk (8000 samples @ 16kHz)

§Performance

Target: <2ms per 500ms chunk
Expected: ~0.16ms (10x headroom)
Optimization: Precomputed coefficients, preallocated buffers

§Example

use speech_prep::preprocessing::{DcHighPassFilter, PreprocessingConfig};

let mut filter = DcHighPassFilter::new(PreprocessingConfig::default())?;
let audio_stream = vec![vec![0.0; 8000], vec![0.1; 8000]];

// Process streaming chunks with state continuity
for chunk in audio_stream {
    let clean = filter.process(&chunk, None)?;
    // No discontinuities at boundaries!
}

Implementations§

Source §

impl DcHighPassFilter

Source

pub fn new(config: PreprocessingConfig) -> Result<Self>

Create a new DC offset removal and high-pass filter.

§Arguments

config - Configuration parameters (cutoff frequency, sample rate, EMA alpha)

§Errors

Returns Error::Configuration if configuration is invalid.

§Example

use speech_prep::preprocessing::{DcHighPassFilter, PreprocessingConfig};

let config = PreprocessingConfig {
    highpass_cutoff_hz: 100.0, // More aggressive
    ..Default::default()
};
let filter = DcHighPassFilter::new(config)?;

Source

pub fn process( &mut self, samples: &[f32], vad_context: Option<&VadContext>, ) -> Result<Vec<f32>>

Process audio samples with DC removal and high-pass filtering.

§Arguments

samples - Input audio samples (typically 500ms chunk = 8000 samples @ 16kHz)
vad_context - Optional VAD state for intelligent DC bias updates

§Returns

Processed audio with DC removed and low frequencies attenuated.

§Performance

Expected: ~0.16ms for 8000 samples (10x better than <2ms target)
Complexity: O(n) where n = samples.len()

§Example

use speech_prep::preprocessing::{DcHighPassFilter, PreprocessingConfig, VadContext};

let mut filter = DcHighPassFilter::new(PreprocessingConfig::default())?;

// Chunk 1
let chunk1 = vec![0.1, 0.2, -0.1, 0.15];
let output1 = filter.process(&chunk1, None)?;

// Chunk 2 (state preserved from chunk1 - no discontinuity!)
let chunk2 = vec![0.2, 0.1, 0.3, 0.0];
let output2 = filter.process(&chunk2, None)?;

Source

pub fn reset(&mut self)

Reset filter state for new audio stream.

Clears filter history (x1, x2, y1, y2) and DC bias estimate. Use this when starting a new, independent audio stream.

§Example

use speech_prep::preprocessing::{DcHighPassFilter, PreprocessingConfig};

let mut filter = DcHighPassFilter::new(PreprocessingConfig::default())?;
let audio_stream_1 = vec![0.0; 8000];
let audio_stream_2 = vec![0.2; 8000];

// Process stream 1
filter.process(&audio_stream_1, None)?;

// Switch to unrelated stream 2 - reset state
filter.reset();
filter.process(&audio_stream_2, None)?;