WebAssembly exmaple for speaker diarization (#1411)

2024-10-10 22:14:45 +08:00
parent 67349b52f2
commit 1d061df355
37 changed files with 1008 additions and 24 deletions
--- a/wasm/CMakeLists.txt
+++ b/wasm/CMakeLists.txt
@@ -18,6 +18,10 @@ if(SHERPA_ONNX_ENABLE_WASM_VAD_ASR)
  add_subdirectory(vad-asr)
 endif()

+if(SHERPA_ONNX_ENABLE_WASM_SPEAKER_DIARIZATION)
+  add_subdirectory(speaker-diarization)
+endif()
+
 if(SHERPA_ONNX_ENABLE_WASM_NODEJS)
  add_subdirectory(nodejs)
 endif()
--- a/wasm/speaker-diarization/CMakeLists.txt
+++ b/wasm/speaker-diarization/CMakeLists.txt
@@ -0,0 +1,61 @@
+if(NOT $ENV{SHERPA_ONNX_IS_USING_BUILD_WASM_SH})
+  message(FATAL_ERROR "Please use ./build-wasm-simd-speaker-diarization.sh to build for WASM for speaker diarization")
+endif()
+
+if(NOT EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/assets/segmentation.onnx" OR NOT EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/assets/embedding.onnx")
+  message(FATAL_ERROR "Please read ${CMAKE_CURRENT_SOURCE_DIR}/assets/README.md before you continue")
+endif()
+
+set(exported_functions
+  MyPrint
+  SherpaOnnxCreateOfflineSpeakerDiarization
+  SherpaOnnxDestroyOfflineSpeakerDiarization
+  SherpaOnnxOfflineSpeakerDiarizationDestroyResult
+  SherpaOnnxOfflineSpeakerDiarizationDestroySegment
+  SherpaOnnxOfflineSpeakerDiarizationGetSampleRate
+  SherpaOnnxOfflineSpeakerDiarizationProcess
+  SherpaOnnxOfflineSpeakerDiarizationProcessWithCallback
+  SherpaOnnxOfflineSpeakerDiarizationResultGetNumSegments
+  SherpaOnnxOfflineSpeakerDiarizationResultSortByStartTime
+  SherpaOnnxOfflineSpeakerDiarizationSetConfig
+)
+set(mangled_exported_functions)
+foreach(x IN LISTS exported_functions)
+  list(APPEND mangled_exported_functions "_${x}")
+endforeach()
+list(JOIN mangled_exported_functions "," all_exported_functions)
+
+
+include_directories(${CMAKE_SOURCE_DIR})
+set(MY_FLAGS " -s FORCE_FILESYSTEM=1 -s INITIAL_MEMORY=512MB -s ALLOW_MEMORY_GROWTH=1")
+string(APPEND MY_FLAGS " -sSTACK_SIZE=10485760 ") # 10MB
+string(APPEND MY_FLAGS " -sEXPORTED_FUNCTIONS=[_CopyHeap,_malloc,_free,${all_exported_functions}] ")
+string(APPEND MY_FLAGS "--preload-file ${CMAKE_CURRENT_SOURCE_DIR}/assets@. ")
+string(APPEND MY_FLAGS " -sEXPORTED_RUNTIME_METHODS=['ccall','stringToUTF8','setValue','getValue','lengthBytesUTF8','UTF8ToString'] ")
+
+message(STATUS "MY_FLAGS: ${MY_FLAGS}")
+
+set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${MY_FLAGS}")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${MY_FLAGS}")
+set(CMAKE_EXECUTBLE_LINKER_FLAGS "${CMAKE_EXECUTBLE_LINKER_FLAGS} ${MY_FLAGS}")
+
+if (NOT CMAKE_EXECUTABLE_SUFFIX STREQUAL ".js")
+  message(FATAL_ERROR "The default suffix for building executables should be .js!")
+endif()
+# set(CMAKE_EXECUTABLE_SUFFIX ".html")
+
+add_executable(sherpa-onnx-wasm-main-speaker-diarization sherpa-onnx-wasm-main-speaker-diarization.cc)
+target_link_libraries(sherpa-onnx-wasm-main-speaker-diarization sherpa-onnx-c-api)
+install(TARGETS sherpa-onnx-wasm-main-speaker-diarization DESTINATION bin/wasm/speaker-diarization)
+
+install(
+  FILES
+    "$<TARGET_FILE_DIR:sherpa-onnx-wasm-main-speaker-diarization>/sherpa-onnx-wasm-main-speaker-diarization.js"
+    "index.html"
+    "sherpa-onnx-speaker-diarization.js"
+    "app-speaker-diarization.js"
+    "$<TARGET_FILE_DIR:sherpa-onnx-wasm-main-speaker-diarization>/sherpa-onnx-wasm-main-speaker-diarization.wasm"
+    "$<TARGET_FILE_DIR:sherpa-onnx-wasm-main-speaker-diarization>/sherpa-onnx-wasm-main-speaker-diarization.data"
+  DESTINATION
+    bin/wasm/speaker-diarization
+)
--- a/wasm/speaker-diarization/app-speaker-diarization.js
+++ b/wasm/speaker-diarization/app-speaker-diarization.js
@@ -0,0 +1,124 @@
+const startBtn = document.getElementById('startBtn');
+const hint = document.getElementById('hint');
+const numClustersInput = document.getElementById('numClustersInputID');
+const thresholdInput = document.getElementById('thresholdInputID');
+const textArea = document.getElementById('text');
+
+const fileSelectCtrl = document.getElementById('file');
+
+let sd = null;
+let float32Samples = null;
+
+Module = {};
+Module.onRuntimeInitialized = function() {
+  console.log('Model files downloaded!');
+
+  console.log('Initializing speaker diarization ......');
+  sd = createOfflineSpeakerDiarization(Module)
+  console.log('sampleRate', sd.sampleRate);
+
+  hint.innerText =
+      'Initialized! Please select a wave file and click the Start button.';
+
+  fileSelectCtrl.disabled = false;
+};
+
+function onFileChange() {
+  var files = document.getElementById('file').files;
+
+  if (files.length == 0) {
+    console.log('No file selected');
+    float32Samples = null;
+    startBtn.disabled = true;
+    return;
+  }
+  textArea.value = '';
+
+  console.log('files: ' + files);
+
+  const file = files[0];
+  console.log(file);
+  console.log('file.name ' + file.name);
+  console.log('file.type ' + file.type);
+  console.log('file.size ' + file.size);
+
+  let audioCtx = new AudioContext({sampleRate: sd.sampleRate});
+
+  let reader = new FileReader();
+  reader.onload = function() {
+    console.log('reading file!');
+    audioCtx.decodeAudioData(reader.result, decodedDone);
+  };
+
+  function decodedDone(decoded) {
+    let typedArray = new Float32Array(decoded.length);
+    float32Samples = decoded.getChannelData(0);
+
+    startBtn.disabled = false;
+  }
+
+  reader.readAsArrayBuffer(file);
+}
+
+startBtn.onclick = function() {
+  textArea.value = '';
+  if (float32Samples == null) {
+    alert('Empty audio samples!');
+
+    startBtn.disabled = true;
+    return;
+  }
+
+  let numClusters = numClustersInput.value;
+  if (numClusters.trim().length == 0) {
+    alert(
+        'Please provide numClusters. Use -1 if you are not sure how many speakers are there');
+    return;
+  }
+
+  if (!numClusters.match(/^\d+$/)) {
+    alert(`number of clusters ${
+        numClusters} is not an integer .\nPlease enter an integer`);
+    return;
+  }
+  numClusters = parseInt(numClusters, 10);
+  if (numClusters < -1) {
+    alert(`Number of clusters should be >= -1`);
+    return;
+  }
+
+  let threshold = 0.5;
+  if (numClusters <= 0) {
+    threshold = thresholdInput.value;
+    if (threshold.trim().length == 0) {
+      alert('Please provide a threshold.');
+      return;
+    }
+
+    threshold = parseFloat(threshold);
+    if (threshold < 0) {
+      alert(`Pleaser enter a positive threshold`);
+      return;
+    }
+  }
+
+  let config = sd.config
+  config.clustering = {numClusters: numClusters, threshold: threshold};
+  sd.setConfig(config);
+  let segments = sd.process(float32Samples);
+  if (segments == null) {
+    textArea.value = 'No speakers detected';
+    return
+  }
+
+  let s = '';
+  let sep = '';
+
+  for (seg of segments) {
+    // clang-format off
+    s += sep + `${seg.start.toFixed(2)} -- ${seg.end.toFixed(2)} speaker_${seg.speaker}`
+    // clang-format on
+    sep = '\n';
+  }
+  textArea.value = s;
+}
--- a/wasm/speaker-diarization/assets/README.md
+++ b/wasm/speaker-diarization/assets/README.md
@@ -0,0 +1,30 @@
+# Introduction
+
+Please refer to
+https://github.com/k2-fsa/sherpa-onnx/releases/tag/speaker-segmentation-models
+to download a speaker segmentation model
+and
+refer to
+https://github.com/k2-fsa/sherpa-onnx/releases/tag/speaker-recongition-models
+to download a speaker embedding extraction model.
+
+Remember to rename the downloaded files.
+
+The following is an example.
+
+
+```bash
+cd wasm/speaker-diarization/assets/
+
+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
+cp sherpa-onnx-pyannote-segmentation-3-0/model.onnx ./segmentation.onnx
+rm -rf sherpa-onnx-pyannote-segmentation-3-0
+
+
+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
+mv 3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx ./embedding.onnx
+
+
+```
--- a/wasm/speaker-diarization/index.html
+++ b/wasm/speaker-diarization/index.html
@@ -0,0 +1,48 @@
+<html lang="en">
+
+<head>
+  <meta charset="utf-8">
+  <meta name="viewport" content="width=device-width" />
+  <title>Next-gen Kaldi WebAssembly with sherpa-onnx for Speaker Diarization</title>
+  <style>
+    h1,div {
+      text-align: center;
+    }
+    textarea {
+      width:100%;
+    }
+  </style>
+</head>
+
+<body>
+  <h1>
+    Next-gen Kaldi + WebAssembly<br/>
+    Speaker Diarization <br> with <a href="https://github.com/k2-fsa/sherpa-onnx">sherpa-onnx</a>
+  </h1>
+  <div>
+    <span id="hint">Loading model ... ...</span>
+    <br/>
+    <br/>
+    <label for="avatar">Choose a wav file:</label>
+    <input type="file" id="file" accept=".wav" onchange="onFileChange()" disabled></input>
+    <br/>
+    <br/>
+    <label for="numClusters" id="numClustersID">Number of speakers: </label>
+    <input type="text" id="numClustersInputID" name="numClusters" value="-1" />
+    <br/>
+    <br/>
+    <label for="clusteringThreshold" id="thresholdID">Clustering threshold: </label>
+    <input type="text" id="thresholdInputID" name="clusteringThreshold" value="0.5" />
+    <br/>
+    <br/>
+
+    <textarea id="text" rows="10" placeholder="If you know the actual number of speakers in the input wave file, please provide it via Number of speakers. Otherwise, please leave Number of speakers to -1 and provide Clustering threshold instead. A larger threshold leads to fewer clusters, i.e., fewer speakers; a smaller threshold leads to more clusters, i.e., more speakers."></textarea>
+    <br/>
+    <br/>
+    <button id="startBtn" disabled>Start</button>
+  </div>
+
+  <script src="app-speaker-diarization.js"></script>
+  <script src="sherpa-onnx-speaker-diarization.js"></script>
+  <script src="sherpa-onnx-wasm-main-speaker-diarization.js"></script>
+</body>
--- a/wasm/speaker-diarization/sherpa-onnx-speaker-diarization.js
+++ b/wasm/speaker-diarization/sherpa-onnx-speaker-diarization.js
@@ -0,0 +1,295 @@
+
+function freeConfig(config, Module) {
+  if ('buffer' in config) {
+    Module._free(config.buffer);
+  }
+
+  if ('config' in config) {
+    freeConfig(config.config, Module)
+  }
+
+  if ('segmentation' in config) {
+    freeConfig(config.segmentation, Module)
+  }
+
+  if ('embedding' in config) {
+    freeConfig(config.embedding, Module)
+  }
+
+  if ('clustering' in config) {
+    freeConfig(config.clustering, Module)
+  }
+
+  Module._free(config.ptr);
+}
+
+function initSherpaOnnxOfflineSpeakerSegmentationPyannoteModelConfig(
+    config, Module) {
+  const modelLen = Module.lengthBytesUTF8(config.model || '') + 1;
+  const n = modelLen;
+  const buffer = Module._malloc(n);
+
+  const len = 1 * 4;
+  const ptr = Module._malloc(len);
+
+  let offset = 0;
+  Module.stringToUTF8(config.model || '', buffer + offset, modelLen);
+  offset += modelLen;
+
+  offset = 0;
+  Module.setValue(ptr, buffer + offset, 'i8*');
+
+  return {
+    buffer: buffer, ptr: ptr, len: len,
+  }
+}
+
+function initSherpaOnnxOfflineSpeakerSegmentationModelConfig(config, Module) {
+  if (!('pyannote' in config)) {
+    config.pyannote = {
+      model: '',
+    };
+  }
+
+  const pyannote = initSherpaOnnxOfflineSpeakerSegmentationPyannoteModelConfig(
+      config.pyannote, Module);
+
+  const len = pyannote.len + 3 * 4;
+  const ptr = Module._malloc(len);
+
+  let offset = 0;
+  Module._CopyHeap(pyannote.ptr, pyannote.len, ptr + offset);
+  offset += pyannote.len;
+
+  Module.setValue(ptr + offset, config.numThreads || 1, 'i32');
+  offset += 4;
+
+  Module.setValue(ptr + offset, config.debug || 1, 'i32');
+  offset += 4;
+
+  const providerLen = Module.lengthBytesUTF8(config.provider || 'cpu') + 1;
+  const buffer = Module._malloc(providerLen);
+  Module.stringToUTF8(config.provider || 'cpu', buffer, providerLen);
+  Module.setValue(ptr + offset, buffer, 'i8*');
+
+  return {
+    buffer: buffer,
+    ptr: ptr,
+    len: len,
+    config: pyannote,
+  };
+}
+
+function initSherpaOnnxSpeakerEmbeddingExtractorConfig(config, Module) {
+  const modelLen = Module.lengthBytesUTF8(config.model || '') + 1;
+  const providerLen = Module.lengthBytesUTF8(config.provider || 'cpu') + 1;
+  const n = modelLen + providerLen;
+  const buffer = Module._malloc(n);
+
+  const len = 4 * 4;
+  const ptr = Module._malloc(len);
+
+  let offset = 0;
+  Module.stringToUTF8(config.model || '', buffer + offset, modelLen);
+  offset += modelLen;
+
+  Module.stringToUTF8(config.provider || 'cpu', buffer + offset, providerLen);
+  offset += providerLen;
+
+  offset = 0
+  Module.setValue(ptr + offset, buffer, 'i8*');
+  offset += 4;
+
+  Module.setValue(ptr + offset, config.numThreads || 1, 'i32');
+  offset += 4;
+
+  Module.setValue(ptr + offset, config.debug || 1, 'i32');
+  offset += 4;
+
+  Module.setValue(ptr + offset, buffer + modelLen, 'i8*');
+  offset += 4;
+
+  return {
+    buffer: buffer,
+    ptr: ptr,
+    len: len,
+  };
+}
+
+function initSherpaOnnxFastClusteringConfig(config, Module) {
+  const len = 2 * 4;
+  const ptr = Module._malloc(len);
+
+  let offset = 0;
+  Module.setValue(ptr + offset, config.numClusters || -1, 'i32');
+  offset += 4;
+
+  Module.setValue(ptr + offset, config.threshold || 0.5, 'float');
+  offset += 4;
+
+  return {
+    ptr: ptr,
+    len: len,
+  };
+}
+
+function initSherpaOnnxOfflineSpeakerDiarizationConfig(config, Module) {
+  if (!('segmentation' in config)) {
+    config.segmentation = {
+      pyannote: {model: ''},
+      numThreads: 1,
+      debug: 0,
+      provider: 'cpu',
+    };
+  }
+
+  if (!('embedding' in config)) {
+    config.embedding = {
+      model: '',
+      numThreads: 1,
+      debug: 0,
+      provider: 'cpu',
+    };
+  }
+
+  if (!('clustering' in config)) {
+    config.clustering = {
+      numClusters: -1,
+      threshold: 0.5,
+    };
+  }
+
+  const segmentation = initSherpaOnnxOfflineSpeakerSegmentationModelConfig(
+      config.segmentation, Module);
+
+  const embedding =
+      initSherpaOnnxSpeakerEmbeddingExtractorConfig(config.embedding, Module);
+
+  const clustering =
+      initSherpaOnnxFastClusteringConfig(config.clustering, Module);
+
+  const len = segmentation.len + embedding.len + clustering.len + 2 * 4;
+  const ptr = Module._malloc(len);
+
+  let offset = 0;
+  Module._CopyHeap(segmentation.ptr, segmentation.len, ptr + offset);
+  offset += segmentation.len;
+
+  Module._CopyHeap(embedding.ptr, embedding.len, ptr + offset);
+  offset += embedding.len;
+
+  Module._CopyHeap(clustering.ptr, clustering.len, ptr + offset);
+  offset += clustering.len;
+
+  Module.setValue(ptr + offset, config.minDurationOn || 0.2, 'float');
+  offset += 4;
+
+  Module.setValue(ptr + offset, config.minDurationOff || 0.5, 'float');
+  offset += 4;
+
+  return {
+    ptr: ptr, len: len, segmentation: segmentation, embedding: embedding,
+        clustering: clustering,
+  }
+}
+
+class OfflineSpeakerDiarization {
+  constructor(configObj, Module) {
+    const config =
+        initSherpaOnnxOfflineSpeakerDiarizationConfig(configObj, Module)
+    // Module._MyPrint(config.ptr);
+
+    const handle =
+        Module._SherpaOnnxCreateOfflineSpeakerDiarization(config.ptr);
+
+    freeConfig(config, Module);
+
+    this.handle = handle;
+    this.sampleRate =
+        Module._SherpaOnnxOfflineSpeakerDiarizationGetSampleRate(this.handle);
+    this.Module = Module
+
+                  this.config = configObj;
+  }
+
+  free() {
+    this.Module._SherpaOnnxDestroyOfflineSpeakerDiarization(this.handle);
+    this.handle = 0
+  }
+
+  setConfig(configObj) {
+    if (!('clustering' in configObj)) {
+      return;
+    }
+
+    const config =
+        initSherpaOnnxOfflineSpeakerDiarizationConfig(configObj, this.Module);
+
+    this.Module._SherpaOnnxOfflineSpeakerDiarizationSetConfig(
+        this.handle, config.ptr);
+
+    freeConfig(config, Module);
+
+    this.config.clustering = configObj.clustering;
+  }
+
+  process(samples) {
+    const pointer =
+        this.Module._malloc(samples.length * samples.BYTES_PER_ELEMENT);
+    this.Module.HEAPF32.set(samples, pointer / samples.BYTES_PER_ELEMENT);
+
+    let r = this.Module._SherpaOnnxOfflineSpeakerDiarizationProcess(
+        this.handle, pointer, samples.length);
+    this.Module._free(pointer);
+
+    let numSegments =
+        this.Module._SherpaOnnxOfflineSpeakerDiarizationResultGetNumSegments(r);
+
+    let segments =
+        this.Module._SherpaOnnxOfflineSpeakerDiarizationResultSortByStartTime(
+            r);
+
+    let ans = [];
+
+    let sizeOfSegment = 3 * 4;
+    for (let i = 0; i < numSegments; ++i) {
+      let p = segments + i * sizeOfSegment
+
+      let start = this.Module.HEAPF32[p / 4 + 0];
+      let end = this.Module.HEAPF32[p / 4 + 1];
+      let speaker = this.Module.HEAP32[p / 4 + 2];
+
+      ans.push({start: start, end: end, speaker: speaker});
+    }
+
+    this.Module._SherpaOnnxOfflineSpeakerDiarizationDestroySegment(segments);
+    this.Module._SherpaOnnxOfflineSpeakerDiarizationDestroyResult(r);
+
+    return ans;
+  }
+}
+
+function createOfflineSpeakerDiarization(Module, myConfig) {
+  const config = {
+    segmentation: {
+      pyannote: {model: './segmentation.onnx'},
+    },
+    embedding: {model: './embedding.onnx'},
+    clustering: {numClusters: -1, threshold: 0.5},
+    minDurationOn: 0.3,
+    minDurationOff: 0.5,
+  };
+
+  if (myConfig) {
+    config = myConfig;
+  }
+
+  return new OfflineSpeakerDiarization(config, Module);
+}
+
+if (typeof process == 'object' && typeof process.versions == 'object' &&
+    typeof process.versions.node == 'string') {
+  module.exports = {
+    createOfflineSpeakerDiarization,
+  };
+}
--- a/wasm/speaker-diarization/sherpa-onnx-wasm-main-speaker-diarization.cc
+++ b/wasm/speaker-diarization/sherpa-onnx-wasm-main-speaker-diarization.cc
@@ -0,0 +1,63 @@
+// wasm/sherpa-onnx-wasm-main-speaker-diarization.cc
+//
+// Copyright (c)  2024  Xiaomi Corporation
+#include <stdio.h>
+
+#include <algorithm>
+#include <memory>
+
+#include "sherpa-onnx/c-api/c-api.h"
+
+// see also
+// https://emscripten.org/docs/porting/connecting_cpp_and_javascript/Interacting-with-code.html
+
+extern "C" {
+
+static_assert(sizeof(SherpaOnnxOfflineSpeakerSegmentationPyannoteModelConfig) ==
+                  1 * 4,
+              "");
+
+static_assert(
+    sizeof(SherpaOnnxOfflineSpeakerSegmentationModelConfig) ==
+        sizeof(SherpaOnnxOfflineSpeakerSegmentationPyannoteModelConfig) + 3 * 4,
+    "");
+
+static_assert(sizeof(SherpaOnnxFastClusteringConfig) == 2 * 4, "");
+
+static_assert(sizeof(SherpaOnnxSpeakerEmbeddingExtractorConfig) == 4 * 4, "");
+
+static_assert(sizeof(SherpaOnnxOfflineSpeakerDiarizationConfig) ==
+                  sizeof(SherpaOnnxOfflineSpeakerSegmentationModelConfig) +
+                      sizeof(SherpaOnnxSpeakerEmbeddingExtractorConfig) +
+                      sizeof(SherpaOnnxFastClusteringConfig) + 2 * 4,
+              "");
+
+void MyPrint(const SherpaOnnxOfflineSpeakerDiarizationConfig *sd_config) {
+  const auto &segmentation = sd_config->segmentation;
+  const auto &embedding = sd_config->embedding;
+  const auto &clustering = sd_config->clustering;
+
+  fprintf(stdout, "----------segmentation config----------\n");
+  fprintf(stdout, "pyannote model: %s\n", segmentation.pyannote.model);
+  fprintf(stdout, "num threads: %d\n", segmentation.num_threads);
+  fprintf(stdout, "debug: %d\n", segmentation.debug);
+  fprintf(stdout, "provider: %s\n", segmentation.provider);
+
+  fprintf(stdout, "----------embedding config----------\n");
+  fprintf(stdout, "model: %s\n", embedding.model);
+  fprintf(stdout, "num threads: %d\n", embedding.num_threads);
+  fprintf(stdout, "debug: %d\n", embedding.debug);
+  fprintf(stdout, "provider: %s\n", embedding.provider);
+
+  fprintf(stdout, "----------clustering config----------\n");
+  fprintf(stdout, "num_clusters: %d\n", clustering.num_clusters);
+  fprintf(stdout, "threshold: %.3f\n", clustering.threshold);
+
+  fprintf(stdout, "min_duration_on: %.3f\n", sd_config->min_duration_on);
+  fprintf(stdout, "min_duration_off: %.3f\n", sd_config->min_duration_off);
+}
+
+void CopyHeap(const char *src, int32_t num_bytes, char *dst) {
+  std::copy(src, src + num_bytes, dst);
+}
+}