Add Swift API for ten-vad (#2387)

.github/scripts/test-swift.sh

@@ -71,7 +71,11 @@ curl -SL -O https://huggingface.co/csukuangfj/test-data/resolve/main/Obama.wav
 ls -lh
 popd
 
+./run-generate-subtitles-ten-vad.sh
+rm -rf *.onnx
+
 ./run-generate-subtitles.sh
+rm -rf *.onnx
 
 ls -lh /Users/fangjun/Desktop
 cat /Users/fangjun/Desktop/Obama.srt

swift-api-examples/.gitignore

@@ -1,6 +1,7 @@
 decode-file
 decode-file-non-streaming
 generate-subtitles
+generate-subtitles-ten-vad
 spoken-language-identification
 tts-vits
 vits-vctk

swift-api-examples/SherpaOnnx.swift

@@ -386,6 +386,22 @@ func sherpaOnnxOfflineWhisperModelConfig(
   )
 }
 
+func sherpaOnnxOfflineCanaryModelConfig(
+  encoder: String = "",
+  decoder: String = "",
+  srcLang: String = "en",
+  tgtLang: String = "en",
+  usePnc: Bool = true
+) -> SherpaOnnxOfflineCanaryModelConfig {
+  return SherpaOnnxOfflineCanaryModelConfig(
+    encoder: toCPointer(encoder),
+    decoder: toCPointer(decoder),
+    src_lang: toCPointer(srcLang),
+    tgt_lang: toCPointer(tgtLang),
+    use_pnc: usePnc ? 1 : 0
+  )
+}
+
 func sherpaOnnxOfflineFireRedAsrModelConfig(
   encoder: String = "",
   decoder: String = ""
@@ -459,7 +475,8 @@ func sherpaOnnxOfflineModelConfig(
   fireRedAsr: SherpaOnnxOfflineFireRedAsrModelConfig = sherpaOnnxOfflineFireRedAsrModelConfig(),
   dolphin: SherpaOnnxOfflineDolphinModelConfig = sherpaOnnxOfflineDolphinModelConfig(),
   zipformerCtc: SherpaOnnxOfflineZipformerCtcModelConfig =
-    sherpaOnnxOfflineZipformerCtcModelConfig()
+    sherpaOnnxOfflineZipformerCtcModelConfig(),
+  canary: SherpaOnnxOfflineCanaryModelConfig = sherpaOnnxOfflineCanaryModelConfig()
 ) -> SherpaOnnxOfflineModelConfig {
   return SherpaOnnxOfflineModelConfig(
     transducer: transducer,
@@ -479,7 +496,8 @@ func sherpaOnnxOfflineModelConfig(
     moonshine: moonshine,
     fire_red_asr: fireRedAsr,
     dolphin: dolphin,
-    zipformer_ctc: zipformerCtc
+    zipformer_ctc: zipformerCtc,
+    canary: canary
   )
 }
 
@@ -607,10 +625,14 @@ class SherpaOnnxOfflineRecognizer {
 
     return SherpaOnnxOfflineRecongitionResult(result: result)
   }
+
+  func setConfig(config: UnsafePointer<SherpaOnnxOfflineRecognizerConfig>!) {
+    SherpaOnnxOfflineRecognizerSetConfig(recognizer, config)
+  }
 }
 
 func sherpaOnnxSileroVadModelConfig(
-  model: String,
+  model: String = "",
   threshold: Float = 0.5,
   minSilenceDuration: Float = 0.25,
   minSpeechDuration: Float = 0.5,
@@ -627,19 +649,39 @@ func sherpaOnnxSileroVadModelConfig(
   )
 }
 
+func sherpaOnnxTenVadModelConfig(
+  model: String = "",
+  threshold: Float = 0.5,
+  minSilenceDuration: Float = 0.25,
+  minSpeechDuration: Float = 0.5,
+  windowSize: Int = 256,
+  maxSpeechDuration: Float = 5.0
+) -> SherpaOnnxTenVadModelConfig {
+  return SherpaOnnxTenVadModelConfig(
+    model: toCPointer(model),
+    threshold: threshold,
+    min_silence_duration: minSilenceDuration,
+    min_speech_duration: minSpeechDuration,
+    window_size: Int32(windowSize),
+    max_speech_duration: maxSpeechDuration
+  )
+}
+
 func sherpaOnnxVadModelConfig(
-  sileroVad: SherpaOnnxSileroVadModelConfig,
+  sileroVad: SherpaOnnxSileroVadModelConfig = sherpaOnnxSileroVadModelConfig(),
   sampleRate: Int32 = 16000,
   numThreads: Int = 1,
   provider: String = "cpu",
-  debug: Int = 0
+  debug: Int = 0,
+  tenVad: SherpaOnnxTenVadModelConfig = sherpaOnnxTenVadModelConfig()
 ) -> SherpaOnnxVadModelConfig {
   return SherpaOnnxVadModelConfig(
     silero_vad: sileroVad,
     sample_rate: sampleRate,
     num_threads: Int32(numThreads),
     provider: toCPointer(provider),
-    debug: Int32(debug)
+    debug: Int32(debug),
+    ten_vad: tenVad
   )
 }
 

swift-api-examples/generate-subtitles.swift

@@ -156,11 +156,35 @@ func run() {
   assert(audioFormat.channelCount == 1)
   assert(audioFormat.commonFormat == AVAudioCommonFormat.pcmFormatFloat32)
 
-  let sileroVadConfig = sherpaOnnxSileroVadModelConfig(
-    model: "./silero_vad.onnx"
-  )
+  var sileroVadConfig = sherpaOnnxSileroVadModelConfig()
+  var tenVadConfig = sherpaOnnxTenVadModelConfig()
+
+  var windowSize = 0
+
+  if FileManager.default.fileExists(atPath: "./silero_vad.onnx") {
+    sileroVadConfig = sherpaOnnxSileroVadModelConfig(
+      model: "./silero_vad.onnx",
+      threshold: 0.25,
+      windowSize: 512
+    )
+    windowSize = 512
+    print("Use silero-vad")
+  } else if FileManager.default.fileExists(atPath: "./ten-vad.onnx") {
+    tenVadConfig = sherpaOnnxTenVadModelConfig(
+      model: "./ten-vad.onnx",
+      threshold: 0.25,
+      windowSize: 256
+    )
+    windowSize = 256
+    print("Use ten-vad")
+  } else {
+    print("Please provide ./silero_vad.onnx or ./ten-vad.onnx")
+    return
+  }
+
+  var vadModelConfig = sherpaOnnxVadModelConfig(
+    sileroVad: sileroVadConfig, tenVad: tenVadConfig)
 
-  var vadModelConfig = sherpaOnnxVadModelConfig(sileroVad: sileroVadConfig)
   let vad = SherpaOnnxVoiceActivityDetectorWrapper(
     config: &vadModelConfig, buffer_size_in_seconds: 120)
 
@@ -170,8 +194,6 @@ func run() {
   try! audioFile.read(into: audioFileBuffer!)
   var array: [Float]! = audioFileBuffer?.array()
 
-  let windowSize = Int(vadModelConfig.silero_vad.window_size)
-
   var segments: [SpeechSegment] = []
 
   for offset in stride(from: 0, to: array.count, by: windowSize) {
@@ -180,7 +202,6 @@ func run() {
   }
 
   vad.flush()
-  var index: Int = 0
   while !vad.isEmpty() {
     let s = vad.front()
     vad.pop()

swift-api-examples/run-generate-subtitles-ten-vad.sh

@@ -0,0 +1,46 @@
+#!/usr/bin/env bash
+
+set -ex
+
+if [ ! -d ../build-swift-macos ]; then
+  echo "Please run ../build-swift-macos.sh first!"
+  exit 1
+fi
+
+if [ ! -d ./sherpa-onnx-whisper-tiny.en ]; then
+  echo "Please download the pre-trained model for testing."
+  echo "You can refer to"
+  echo ""
+  echo "https://k2-fsa.github.io/sherpa/onnx/pretrained_models/whisper/tiny.en.html"
+  echo ""
+  echo "for help"
+
+  wget -q https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-whisper-tiny.en.tar.bz2
+  tar xvf sherpa-onnx-whisper-tiny.en.tar.bz2
+  rm sherpa-onnx-whisper-tiny.en.tar.bz2
+  ls -lh sherpa-onnx-whisper-tiny.en
+fi
+if [ ! -f ./ten-vad.onnx ]; then
+  echo "downloading ten-vad"
+  wget -q https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/ten-vad.onnx
+fi
+
+if [ ! -e ./generate-subtitles-ten-vad ]; then
+  # Note: We use -lc++ to link against libc++ instead of libstdc++
+  swiftc \
+    -lc++ \
+    -I ../build-swift-macos/install/include \
+    -import-objc-header ./SherpaOnnx-Bridging-Header.h \
+    ./generate-subtitles.swift  ./SherpaOnnx.swift \
+    -L ../build-swift-macos/install/lib/ \
+    -l sherpa-onnx \
+    -l onnxruntime \
+    -o generate-subtitles-ten-vad
+
+  strip generate-subtitles-ten-vad
+else
+  echo "./generate-subtitles-ten-vad exists - skip building"
+fi
+
+export DYLD_LIBRARY_PATH=$PWD/../build-swift-macos/install/lib:$DYLD_LIBRARY_PATH
+./generate-subtitles-ten-vad