Struct EmbeddingExtractor

Source

pub struct EmbeddingExtractor { /* private fields */ }

Implementations§

Source §

impl EmbeddingExtractor

Source

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

Examples found in repository ?

examples/infinite.rs (line 43)

35fn main() -> Result<(), eyre::Report> {
36    let audio_path = std::env::args().nth(1).expect("Please specify audio file");
37    let search_threshold = 0.5;
38
39    let embedding_model_path = "wespeaker_en_voxceleb_CAM++.onnx";
40    let segmentation_model_path = "segmentation-3.0.onnx";
41
42    let (samples, sample_rate) = pyannote_rs::read_wav(&audio_path)?;
43    let mut embedding_extractor = EmbeddingExtractor::new(embedding_model_path)?;
44    let mut embedding_manager = EmbeddingManager::new(usize::MAX);
45
46    let segments = pyannote_rs::get_segments(&samples, sample_rate, segmentation_model_path)?;
47
48    for segment in segments {
49        if let Ok(segment) = segment {
50            if let Err(error) = process_segment(
51                segment,
52                &mut embedding_extractor,
53                &mut embedding_manager,
54                search_threshold,
55            ) {
56                eprintln!("Error processing segment: {:?}", error);
57            }
58        } else if let Err(error) = segment {
59            eprintln!("Failed to process segment: {:?}", error);
60        }
61    }
62
63    Ok(())
64}

More examples

Hide additional examples

examples/max_speakers.rs (line 8)

3fn main() {
4    let audio_path = std::env::args().nth(1).expect("Please specify audio file");
5    let (samples, sample_rate) = pyannote_rs::read_wav(&audio_path).unwrap();
6    let max_speakers = 6;
7
8    let mut extractor = EmbeddingExtractor::new("wespeaker_en_voxceleb_CAM++.onnx").unwrap();
9    let mut manager = EmbeddingManager::new(6);
10
11    let segments =
12        pyannote_rs::get_segments(&samples, sample_rate, "segmentation-3.0.onnx").unwrap();
13
14    for segment in segments {
15        match segment {
16            Ok(segment) => {
17                if let Ok(embedding) = extractor.compute(&segment.samples) {
18                    let speaker = if manager.get_all_speakers().len() == max_speakers {
19                        manager
20                            .get_best_speaker_match(embedding.collect())
21                            .map(|s| s.to_string())
22                            .unwrap_or("?".into())
23                    } else {
24                        manager
25                            .search_speaker(embedding.collect(), 0.5)
26                            .map(|s| s.to_string())
27                            .unwrap_or("?".into())
28                    };
29                    println!(
30                        "start = {:.2}, end = {:.2}, speaker = {}",
31                        segment.start, segment.end, speaker
32                    );
33                } else {
34                    println!(
35                        "start = {:.2}, end = {:.2}, speaker = ?",
36                        segment.start, segment.end
37                    );
38                }
39            }
40            Err(error) => eprintln!("Failed to process segment: {:?}", error),
41        }
42    }
43}

Source

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

Examples found in repository ?

examples/infinite.rs (line 17)

10fn process_segment(
11    segment: pyannote_rs::Segment,
12    embedding_extractor: &mut EmbeddingExtractor,
13    embedding_manager: &mut EmbeddingManager,
14    search_threshold: f32,
15) -> Result<(), eyre::Report> {
16    let embedding_result: Vec<f32> = embedding_extractor
17        .compute(&segment.samples)
18        .unwrap()
19        .collect();
20
21    let speaker = embedding_manager
22        .search_speaker(embedding_result.clone(), search_threshold)
23        .ok_or_else(|| embedding_manager.search_speaker(embedding_result, 0.0)) // Ensure always to return speaker
24        .map(|r| r.to_string())
25        .unwrap_or("?".into());
26
27    println!(
28        "start = {:.2}, end = {:.2}, speaker = {}",
29        segment.start, segment.end, speaker
30    );
31
32    Ok(())
33}

More examples

Hide additional examples

examples/max_speakers.rs (line 17)

3fn main() {
4    let audio_path = std::env::args().nth(1).expect("Please specify audio file");
5    let (samples, sample_rate) = pyannote_rs::read_wav(&audio_path).unwrap();
6    let max_speakers = 6;
7
8    let mut extractor = EmbeddingExtractor::new("wespeaker_en_voxceleb_CAM++.onnx").unwrap();
9    let mut manager = EmbeddingManager::new(6);
10
11    let segments =
12        pyannote_rs::get_segments(&samples, sample_rate, "segmentation-3.0.onnx").unwrap();
13
14    for segment in segments {
15        match segment {
16            Ok(segment) => {
17                if let Ok(embedding) = extractor.compute(&segment.samples) {
18                    let speaker = if manager.get_all_speakers().len() == max_speakers {
19                        manager
20                            .get_best_speaker_match(embedding.collect())
21                            .map(|s| s.to_string())
22                            .unwrap_or("?".into())
23                    } else {
24                        manager
25                            .search_speaker(embedding.collect(), 0.5)
26                            .map(|s| s.to_string())
27                            .unwrap_or("?".into())
28                    };
29                    println!(
30                        "start = {:.2}, end = {:.2}, speaker = {}",
31                        segment.start, segment.end, speaker
32                    );
33                } else {
34                    println!(
35                        "start = {:.2}, end = {:.2}, speaker = ?",
36                        segment.start, segment.end
37                    );
38                }
39            }
40            Err(error) => eprintln!("Failed to process segment: {:?}", error),
41        }
42    }
43}