Add Rust API for speaker diarization

.github/scripts/test-rust.sh

@@ -18,6 +18,10 @@ rm -rf sr-data
 ./run-speaker-embedding-cosine-similarity.sh
 rm -f wespeaker_zh_cnceleb_resnet34.onnx fangjun-sr-1.wav fangjun-sr-2.wav leijun-sr-1.wav
 
+./run-offline-speaker-diarization.sh
+rm -rf sherpa-onnx-pyannote-segmentation-3-0
+rm -f 3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx 0-four-speakers-zh.wav
+
 ./run-vits-en.sh
 rm -rf vits-piper-en_US-amy-low
 

rust-api-examples/README.md

@@ -65,6 +65,7 @@ export RUSTFLAGS="-C link-arg=-Wl,-rpath,$SHERPA_ONNX_LIB_DIR"
 | 30 | [speaker_embedding_extractor](#example-30-speaker-embedding-extractor) | Compute a speaker embedding from a wave file |
 | 31 | [speaker_embedding_manager](#example-31-speaker-embedding-manager) | Register, search, verify, and remove speakers using embeddings |
 | 32 | [speaker_embedding_cosine_similarity](#example-32-speaker-embedding-cosine-similarity) | Compute cosine similarity from three speaker embeddings |
+| 33 | [offline_speaker_diarization](#example-33-offline-speaker-diarization) | Offline speaker diarization with pyannote segmentation and 3D-Speaker embeddings |
 
 ## Run it
 
@@ -270,3 +271,10 @@ to check the RPATH.
 ```bash
 ./run-speaker-embedding-cosine-similarity.sh
 ```
+
+
+### Example 33: Offline speaker diarization
+
+```bash
+./run-offline-speaker-diarization.sh
+```

rust-api-examples/examples/offline_speaker_diarization.rs

@@ -0,0 +1,46 @@
+use sherpa_onnx::{
+    FastClusteringConfig, OfflineSpeakerDiarization, OfflineSpeakerDiarizationConfig,
+    OfflineSpeakerSegmentationModelConfig, OfflineSpeakerSegmentationPyannoteModelConfig,
+    SpeakerEmbeddingExtractorConfig, Wave,
+};
+
+fn main() {
+    let config = OfflineSpeakerDiarizationConfig {
+        segmentation: OfflineSpeakerSegmentationModelConfig {
+            pyannote: OfflineSpeakerSegmentationPyannoteModelConfig {
+                model: Some("./sherpa-onnx-pyannote-segmentation-3-0/model.onnx".into()),
+            },
+            ..Default::default()
+        },
+        embedding: SpeakerEmbeddingExtractorConfig {
+            model: Some("./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx".into()),
+            ..Default::default()
+        },
+        clustering: FastClusteringConfig {
+            num_clusters: 4,
+            ..Default::default()
+        },
+        ..Default::default()
+    };
+
+    let sd = OfflineSpeakerDiarization::create(&config)
+        .expect("Failed to initialize offline speaker diarization");
+
+    let wave = Wave::read("./0-four-speakers-zh.wav").expect("Failed to read wave");
+
+    assert_eq!(
+        sd.sample_rate(),
+        wave.sample_rate(),
+        "Unexpected sample rate"
+    );
+
+    let result = sd
+        .process(wave.samples())
+        .expect("Failed to do speaker diarization");
+    println!("Number of speakers: {}", result.num_speakers());
+    println!("Number of segments: {}", result.num_segments());
+
+    for s in result.sort_by_start_time() {
+        println!("{:.3} -- {:.3} speaker_{:02}", s.start, s.end, s.speaker);
+    }
+}

rust-api-examples/run-offline-speaker-diarization.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+set -ex
+
+if [ ! -f ./sherpa-onnx-pyannote-segmentation-3-0/model.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-segmentation-models/sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+  tar xvf sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+  rm sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+fi
+
+if [ ! -f ./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
+fi
+
+if [ ! -f ./0-four-speakers-zh.wav ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-segmentation-models/0-four-speakers-zh.wav
+fi
+
+cargo run --example offline_speaker_diarization

sherpa-onnx/rust/sherpa-onnx-sys/src/lib.rs

@@ -13,6 +13,7 @@ extern "C" {
 
 pub mod audio_tagging;
 pub mod offline_asr;
+pub mod offline_speaker_diarization;
 pub mod online_asr;
 pub mod online_punctuation;
 pub mod speaker_embedding;
@@ -23,6 +24,7 @@ pub mod wave;
 
 pub use audio_tagging::*;
 pub use offline_asr::*;
+pub use offline_speaker_diarization::*;
 pub use online_asr::*;
 pub use online_punctuation::*;
 pub use speaker_embedding::*;

sherpa-onnx/rust/sherpa-onnx-sys/src/offline_speaker_diarization.rs

@@ -0,0 +1,98 @@
+#![allow(non_camel_case_types)]
+#![allow(non_snake_case)]
+#![allow(non_upper_case_globals)]
+
+use std::os::raw::{c_char, c_float};
+
+#[repr(C)]
+#[derive(Debug, Copy, Clone)]
+pub struct OfflineSpeakerSegmentationPyannoteModelConfig {
+    pub model: *const c_char,
+}
+
+#[repr(C)]
+#[derive(Debug, Copy, Clone)]
+pub struct OfflineSpeakerSegmentationModelConfig {
+    pub pyannote: OfflineSpeakerSegmentationPyannoteModelConfig,
+    pub num_threads: i32,
+    pub debug: i32,
+    pub provider: *const c_char,
+}
+
+#[repr(C)]
+#[derive(Debug, Copy, Clone)]
+pub struct FastClusteringConfig {
+    pub num_clusters: i32,
+    pub threshold: c_float,
+}
+
+#[repr(C)]
+#[derive(Debug, Copy, Clone)]
+pub struct OfflineSpeakerDiarizationConfig {
+    pub segmentation: OfflineSpeakerSegmentationModelConfig,
+    pub embedding: crate::speaker_embedding::SpeakerEmbeddingExtractorConfig,
+    pub clustering: FastClusteringConfig,
+    pub min_duration_on: c_float,
+    pub min_duration_off: c_float,
+}
+
+#[repr(C)]
+pub struct OfflineSpeakerDiarization {
+    _private: [u8; 0],
+}
+
+#[repr(C)]
+pub struct OfflineSpeakerDiarizationResult {
+    _private: [u8; 0],
+}
+
+#[repr(C)]
+#[derive(Debug, Copy, Clone)]
+pub struct OfflineSpeakerDiarizationSegment {
+    pub start: c_float,
+    pub end: c_float,
+    pub speaker: i32,
+}
+
+extern "C" {
+    pub fn SherpaOnnxCreateOfflineSpeakerDiarization(
+        config: *const OfflineSpeakerDiarizationConfig,
+    ) -> *const OfflineSpeakerDiarization;
+
+    pub fn SherpaOnnxDestroyOfflineSpeakerDiarization(sd: *const OfflineSpeakerDiarization);
+
+    pub fn SherpaOnnxOfflineSpeakerDiarizationGetSampleRate(
+        sd: *const OfflineSpeakerDiarization,
+    ) -> i32;
+
+    pub fn SherpaOnnxOfflineSpeakerDiarizationSetConfig(
+        sd: *const OfflineSpeakerDiarization,
+        config: *const OfflineSpeakerDiarizationConfig,
+    );
+
+    pub fn SherpaOnnxOfflineSpeakerDiarizationResultGetNumSpeakers(
+        r: *const OfflineSpeakerDiarizationResult,
+    ) -> i32;
+
+    pub fn SherpaOnnxOfflineSpeakerDiarizationResultGetNumSegments(
+        r: *const OfflineSpeakerDiarizationResult,
+    ) -> i32;
+
+    pub fn SherpaOnnxOfflineSpeakerDiarizationResultSortByStartTime(
+        r: *const OfflineSpeakerDiarizationResult,
+    ) -> *const OfflineSpeakerDiarizationSegment;
+
+    pub fn SherpaOnnxOfflineSpeakerDiarizationDestroySegment(
+        s: *const OfflineSpeakerDiarizationSegment,
+    );
+
+    pub fn SherpaOnnxOfflineSpeakerDiarizationProcess(
+        sd: *const OfflineSpeakerDiarization,
+        samples: *const c_float,
+        n: i32,
+    ) -> *const OfflineSpeakerDiarizationResult;
+
+    pub fn SherpaOnnxOfflineSpeakerDiarizationDestroyResult(
+        r: *const OfflineSpeakerDiarizationResult,
+    );
+}

sherpa-onnx/rust/sherpa-onnx/src/lib.rs

@@ -1,6 +1,7 @@
 mod audio_tagging;
 mod display;
 mod offline_asr;
+mod offline_speaker_diarization;
 mod offline_speech_denoiser;
 mod online_asr;
 mod online_punctuation;
@@ -15,6 +16,7 @@ mod wave;
 pub use audio_tagging::*;
 pub use display::*;
 pub use offline_asr::*;
+pub use offline_speaker_diarization::*;
 pub use offline_speech_denoiser::*;
 pub use online_asr::*;
 pub use online_punctuation::*;