Crate polyvoice 

polyvoice

Speaker diarization library for Rust — online (streaming) and offline (file-based), ONNX-powered, and ecosystem-agnostic.

Designed to be embedded into any Rust application that needs to answer the question “who spoke when?”.

Quick start

Build a diarization pipeline using Pipeline and ModelRegistry. See the pipeline module for details.

Module organization

polyvoice carries two parallel module families from an in-progress migration to a trait-based v1.0 architecture. This is deliberate (shared math, a compile-time feature guard), not accidental duplication:

• v1.0 trait-based (current architecture): embedder (the migration target trait), clusterer, segmentation, resegmentation, silero_vad, and pipeline_v2 (experimental — see its README).
• Legacy: embedding, ecapa, onnx are #[deprecated] — migrate to the embedder trait. cluster and vad are the legacy clustering/VAD surfaces.
• Pipeline status (note the inversion): the legacy pipeline is the validated default for the CLI and Python bindings; pipeline_v2 is experimental (opt-in via --v2), reverted from default after the 0.6.1 long-form DER regression.
• Shared math, reused by both families: ahc, kmeans, spectral, features, der, utils.

Re-exports

pub use features::FbankConfig;

pub use features::FbankExtractor;

pub use utils::merge_segments;

pub use segmentation::AggregationConfig;

pub use segmentation::Aggregator;

pub use segmentation::FrameLabel;

pub use segmentation::MIN_AUDIO_SAMPLES;

pub use segmentation::PowersetClass;

pub use segmentation::PowersetDecoder;

pub use segmentation::RawSegment;

pub use segmentation::SegmentationError;

pub use segmentation::Segmenter;

pub use segmentation::WindowOutput;

pub use segmentation::PowersetConfig;

pub use segmentation::PowersetSegmenter;

pub use embedder::Embedder;

pub use embedder::EmbedderError;

pub use embedder::EmbedderPool;

pub use embedder::apply_overlap_mask;

pub use embedder::CamPlusPlusExtractor;

pub use embedder::ResNet34Adapter;

pub use clusterer::AhcClusterer;

pub use clusterer::Clusterer;

pub use clusterer::ClustererError;

pub use clusterer::NmeScClusterer;

pub use resegmentation::OverlapRegionInput;

pub use resegmentation::OverlapResegmenter;

pub use resegmentation::ResegmentError;

pub use resegmentation::ResegmentInputs;

pub use resegmentation::Resegmenter;

pub use resegmentation::SpeakerCentroid;

pub use resegmentation::compute_centroids;

pub use resegmentation::extract_overlap_time_ranges;

pub use pipeline::Pipeline;

pub use pipeline::PipelineError;

pub use vad::EnergyVad;

pub use vad::VadConfig;

pub use vad::VadError;

pub use vad::VoiceActivityDetector;

pub use vad::segment_speech;

pub use silero_vad::SileroVad;

pub use onnx::OnnxEmbeddingExtractor;Deprecated

pub use cluster::SpeakerCluster;

pub use embedding::DummyExtractor;Deprecated

pub use embedding::EmbeddingError;Deprecated

pub use embedding::EmbeddingExtractor;Deprecated

pub use models::ModelRegistry;

pub use models::ProfileModels;

pub use models::RegistryError;

pub use overlap::OverlapRegion;

pub use overlap::detect_overlaps;

pub use types::ClusterConfig;

pub use types::Confidence;

pub use types::DiarizationConfig;

pub use types::DiarizationResult;

pub use types::Profile;

pub use types::SampleRate;

pub use types::Seconds;

pub use types::Segment;

pub use types::SpeakerId;

pub use types::SpeakerIdRemap;

pub use types::SpeakerTurn;

pub use types::TimeRange;

pub use types::WordAlignment;

pub use types::remap_segments;

pub use types::remap_turns;

pub use window::WindowBuffer;

pub use window::WindowIter;

pub use ecapa::FbankOnnxExtractor;Deprecated

Modules

ahc

Agglomerative Hierarchical Clustering (AHC) for speaker diarization.

cluster

Speaker clustering with online incremental centroid updates.

clusterer

v1.0 Clusterer trait + concrete clusterers (NME-SC, AHC).

der

Diarization Error Rate (DER) computation.

ecapa

ONNX speaker embedding extractor (WeSpeaker, ECAPA-TDNN, etc.).

embedder

v1.0 Embedder trait + concrete extractors (CAM++, ResNet34) + pool + overlap-mask helper.

embedding

Speaker embedding extraction trait.

features

Log-mel filterbank (fbank) feature extraction for speaker embeddings.

kmeans

K-Means++ clustering with automatic k selection via silhouette score.

models

Model registry — manifest-driven downloads with SHA-256 verification.

onnx

ONNX-based speaker embedding extractor with a session pool.

overlap

Overlap detection: identify frames where multiple speakers may be active.

pipeline

High-level diarization pipeline.

pipeline_v2

M6a — additive polyvoice::pipeline_v2 module.

resegmentation

v1.0 OverlapResegmenter — overlap-aware post-clustering pass.

rttm

RTTM (Rich Transcription Time Marked) parser and writer.

segmentation

Speaker segmentation: powerset-classifier + sliding-window aggregator.

silero_vad

Silero VAD v5 ONNX integration.

spectral

Spectral clustering for speaker diarization.

streaming

Real-time streaming diarization pipeline.

types

Core types for speaker diarization.

utils

Math utilities for diarization.

vad

Voice Activity Detection (VAD) trait and utilities.

wav

WAV file I/O via the hound crate.

window

Sliding-window utilities for batch and streaming pipelines.