Struct LocalVadDetector

pub struct LocalVadDetector { /* private fields */ }

Local voice activity detector.

Provides voice activity detection using local processing without external API calls. Uses the voice_activity_detector crate for speech detection and analysis.

Implementations

impl LocalVadDetector

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

Create a new local VAD detector.

Arguments

• config - VAD configuration parameters

Returns

A new LocalVadDetector instance

Errors

Returns an error if the audio processor cannot be initialized

pub async fn detect_speech(&self, audio_path: &Path) -> Result<VadResult>

Detect speech activity in an audio file.

Processes the entire audio file to identify speech segments with timestamps and confidence scores.

Arguments

• audio_path - Path to the audio file to analyze

Returns

VAD analysis results including speech segments and metadata

Errors

Returns an error if:

• Audio file cannot be loaded
• VAD processing fails
• Audio format is unsupported

pub fn calculate_chunk_size(&self, sample_rate: u32) -> usize

Dynamically calculates the optimal VAD chunk size for a given audio sample rate.

This function selects a chunk size (in samples) that is compatible with the VAD model’s requirements and recommended for common sample rates. For 8000 Hz and 16000 Hz, it uses 512 samples by default, which is within the recommended range (512, 768, or 1024). For other sample rates, it uses a 30 ms window as the baseline, with a minimum of 1024 samples. The function also ensures that the chunk size always satisfies the model’s constraint: sample_rate <= 31.25 * chunk_size.

Arguments

• sample_rate: The audio sample rate in Hz (e.g., 16000 for 16kHz audio)

Returns

The chunk size in number of samples, selected for optimal model compatibility.

Examples

Basic usage:

use subx_cli::services::vad::LocalVadDetector;
let detector = LocalVadDetector::new(Default::default()).unwrap();
let chunk_size = detector.calculate_chunk_size(16000);
assert_eq!(chunk_size, 512);

Model Constraint

The returned chunk size always satisfies: sample_rate <= 31.25 * chunk_size.