Module utils 

Math utilities for diarization.

Shared vector math (cosine similarity, L2 normalization, pairwise similarity matrices, mean centroids, segment merging) used by clustering, embedding, and overlap modules. See cosine_similarity.

Functions

cosine_similarity

{ true } pub fn cosine_similarity(a: &f32, b: &f32) -> f32 { ret >= -1.0 && ret <= 1.0 } Compute cosine similarity between two vectors.

cosine_similarity_f32_f64

{ true } pub fn cosine_similarity_f32_f64(a: &f32, b: &f64) -> f32 { ret >= -1.0 && ret <= 1.0 } Compute cosine similarity between an f32 slice and an f64 slice.

l2_normalize

{ true } pub fn l2_normalize(vec: &mut f32) { true } L2-normalize a vector in-place.

mean_vector

{ true } pub fn mean_vector(vectors: &[Vec<f32>]) -> Option<Vec<f32>> { ret.as_ref().map_or(true, |v| vectors.iter().all(|u| u.len() == v.len())) } Compute the element-wise mean of a list of vectors.

merge_segments

{ true } pub fn merge_segments(segments: Vec<Segment>, max_gap_secs: f64) -> Vec<Segment> { ret.len() <= segments.len() } Merge adjacent segments with the same speaker if the gap between them is less than max_gap_secs.