polyvoice 0.6.5

Speaker diarization library for Rust — online and offline, ONNX-powered, ecosystem-agnostic
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36

# src/segmentation

## Purpose

Speaker segmentation algorithms: powerset decoder, frame aggregation with
Hungarian assignment, and segmenter trait.

## Surfaces

- `Segmenter` trait
- `PowersetSegmenter` (requires `onnx`)
- `PowersetDecoder`
- `Aggregator`
- `FrameLabel`
- `MIN_AUDIO_SAMPLES`

## Dependencies

- `types` — Confidence, TimeRange

## Invariants

- PowersetDecoder is deterministic for identical logits.
- Hungarian assignment minimizes total cost across windows.

## Verification

```bash
cargo test --lib segmentation
cargo test --test chaos_test --features onnx,download
```

## Notes

- Pure-Rust core (decoder, aggregator, Hungarian) compiles to wasm32.
- ONNX-backed PowersetSegmenter requires `onnx` feature.