Struct pyannote_rs::EmbeddingExtractor

pub struct EmbeddingExtractor { /* private fields */ }

Implementations

impl EmbeddingExtractor

pub fn new<P: AsRef<Path>>(model_path: P) -> Result<Self>

Examples found in repository

examples/infinite.rs (line 19)

fn main() {
    let audio_path = std::env::args().nth(1).expect("Please specify audio file");
    let search_threshold = 0.5;

    let embedding_model_path = "wespeaker_en_voxceleb_CAM++.onnx";
    let segmentation_model_path = "segmentation-3.0.onnx";

    let (samples, sample_rate) = pyannote_rs::read_wav(&audio_path).unwrap();
    let mut embedding_extractor = EmbeddingExtractor::new(embedding_model_path).unwrap();
    let mut embedding_manager = EmbeddingManager::new(usize::MAX);

    let segments = pyannote_rs::segment(&samples, sample_rate, segmentation_model_path).unwrap();

    for segment in segments {
        // Compute the embedding result
        let embedding_result: Vec<f32> = match embedding_extractor.compute(&segment.samples) {
            Ok(result) => result.collect(),
            Err(error) => {
                println!(
                    "Error in {:.2}s: {:.2}s: {:?}",
                    segment.start, segment.end, error
                );
                println!(
                    "start = {:.2}, end = {:.2}, speaker = ?",
                    segment.start, segment.end
                );
                continue; // Skip to the next segment
            }
        };

        // Find the speaker
        let speaker = embedding_manager
            .search_speaker(embedding_result.clone(), search_threshold)
            .ok_or_else(|| embedding_manager.search_speaker(embedding_result, 0.0)) // Ensure always to return speaker
            .map(|r| r.to_string())
            .unwrap_or("?".into());

        println!(
            "start = {:.2}, end = {:.2}, speaker = {}",
            segment.start, segment.end, speaker
        );
    }
}

examples/max_speakers.rs (line 20)

fn main() {
    let audio_path = std::env::args().nth(1).expect("Please specify audio file");
    let max_speakers = 6;
    let search_threshold = 0.5;

    let embedding_model_path = "wespeaker_en_voxceleb_CAM++.onnx";
    let segmentation_model_path = "segmentation-3.0.onnx";

    let (samples, sample_rate) = pyannote_rs::read_wav(&audio_path).unwrap();
    let mut embedding_extractor = EmbeddingExtractor::new(embedding_model_path).unwrap();
    let mut embedding_manager = EmbeddingManager::new(max_speakers);

    let segments = pyannote_rs::segment(&samples, sample_rate, segmentation_model_path).unwrap();

    for segment in segments {
        // Compute the embedding result
        let embedding_result = match embedding_extractor.compute(&segment.samples) {
            Ok(result) => result.collect(),
            Err(error) => {
                println!(
                    "Error in {:.2}s: {:.2}s: {:?}",
                    segment.start, segment.end, error
                );
                println!(
                    "start = {:.2}, end = {:.2}, speaker = ?",
                    segment.start, segment.end
                );
                continue; // Skip to the next segment
            }
        };

        // Find the speaker
        let speaker = if embedding_manager.get_all_speakers().len() == max_speakers {
            embedding_manager
                .get_best_speaker_match(embedding_result)
                .map(|r| r.to_string())
                .unwrap_or("?".into())
        } else {
            embedding_manager
                .search_speaker(embedding_result, search_threshold)
                .map(|r| r.to_string())
                .unwrap_or("?".into())
        };

        println!(
            "start = {:.2}, end = {:.2}, speaker = {}",
            segment.start, segment.end, speaker
        );
    }
}

pub fn compute(&mut self, samples: &[i16]) -> Result<impl Iterator<Item = f32>>

Examples found in repository

examples/infinite.rs (line 26)

fn main() {
    let audio_path = std::env::args().nth(1).expect("Please specify audio file");
    let search_threshold = 0.5;

    let embedding_model_path = "wespeaker_en_voxceleb_CAM++.onnx";
    let segmentation_model_path = "segmentation-3.0.onnx";

    let (samples, sample_rate) = pyannote_rs::read_wav(&audio_path).unwrap();
    let mut embedding_extractor = EmbeddingExtractor::new(embedding_model_path).unwrap();
    let mut embedding_manager = EmbeddingManager::new(usize::MAX);

    let segments = pyannote_rs::segment(&samples, sample_rate, segmentation_model_path).unwrap();

    for segment in segments {
        // Compute the embedding result
        let embedding_result: Vec<f32> = match embedding_extractor.compute(&segment.samples) {
            Ok(result) => result.collect(),
            Err(error) => {
                println!(
                    "Error in {:.2}s: {:.2}s: {:?}",
                    segment.start, segment.end, error
                );
                println!(
                    "start = {:.2}, end = {:.2}, speaker = ?",
                    segment.start, segment.end
                );
                continue; // Skip to the next segment
            }
        };

        // Find the speaker
        let speaker = embedding_manager
            .search_speaker(embedding_result.clone(), search_threshold)
            .ok_or_else(|| embedding_manager.search_speaker(embedding_result, 0.0)) // Ensure always to return speaker
            .map(|r| r.to_string())
            .unwrap_or("?".into());

        println!(
            "start = {:.2}, end = {:.2}, speaker = {}",
            segment.start, segment.end, speaker
        );
    }
}

examples/max_speakers.rs (line 27)

fn main() {
    let audio_path = std::env::args().nth(1).expect("Please specify audio file");
    let max_speakers = 6;
    let search_threshold = 0.5;

    let embedding_model_path = "wespeaker_en_voxceleb_CAM++.onnx";
    let segmentation_model_path = "segmentation-3.0.onnx";

    let (samples, sample_rate) = pyannote_rs::read_wav(&audio_path).unwrap();
    let mut embedding_extractor = EmbeddingExtractor::new(embedding_model_path).unwrap();
    let mut embedding_manager = EmbeddingManager::new(max_speakers);

    let segments = pyannote_rs::segment(&samples, sample_rate, segmentation_model_path).unwrap();

    for segment in segments {
        // Compute the embedding result
        let embedding_result = match embedding_extractor.compute(&segment.samples) {
            Ok(result) => result.collect(),
            Err(error) => {
                println!(
                    "Error in {:.2}s: {:.2}s: {:?}",
                    segment.start, segment.end, error
                );
                println!(
                    "start = {:.2}, end = {:.2}, speaker = ?",
                    segment.start, segment.end
                );
                continue; // Skip to the next segment
            }
        };

        // Find the speaker
        let speaker = if embedding_manager.get_all_speakers().len() == max_speakers {
            embedding_manager
                .get_best_speaker_match(embedding_result)
                .map(|r| r.to_string())
                .unwrap_or("?".into())
        } else {
            embedding_manager
                .search_speaker(embedding_result, search_threshold)
                .map(|r| r.to_string())
                .unwrap_or("?".into())
        };

        println!(
            "start = {:.2}, end = {:.2}, speaker = {}",
            segment.start, segment.end, speaker
        );
    }
}

Trait Implementations

impl Debug for EmbeddingExtractor

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

Formats the value using the given formatter.