Add C API for streaming HLG decoding (#734)

swift-api-examples/.gitignore

@@ -7,3 +7,4 @@ vits-vctk
 sherpa-onnx-paraformer-zh-2023-09-14
 !*.sh
 *.bak
+streaming-hlg-decode-file

swift-api-examples/SherpaOnnx.swift

@@ -111,6 +111,15 @@ func sherpaOnnxFeatureConfig(
     feature_dim: Int32(featureDim))
 }
 
+func sherpaOnnxOnlineCtcFstDecoderConfig(
+  graph: String = "",
+  maxActive: Int = 3000
+) -> SherpaOnnxOnlineCtcFstDecoderConfig {
+  return SherpaOnnxOnlineCtcFstDecoderConfig(
+    graph: toCPointer(graph),
+    max_active: Int32(maxActive))
+}
+
 func sherpaOnnxOnlineRecognizerConfig(
   featConfig: SherpaOnnxFeatureConfig,
   modelConfig: SherpaOnnxOnlineModelConfig,
@@ -121,7 +130,8 @@ func sherpaOnnxOnlineRecognizerConfig(
   decodingMethod: String = "greedy_search",
   maxActivePaths: Int = 4,
   hotwordsFile: String = "",
-  hotwordsScore: Float = 1.5
+  hotwordsScore: Float = 1.5,
+  ctcFstDecoderConfig: SherpaOnnxOnlineCtcFstDecoderConfig = sherpaOnnxOnlineCtcFstDecoderConfig()
 ) -> SherpaOnnxOnlineRecognizerConfig {
   return SherpaOnnxOnlineRecognizerConfig(
     feat_config: featConfig,
@@ -133,7 +143,9 @@ func sherpaOnnxOnlineRecognizerConfig(
     rule2_min_trailing_silence: rule2MinTrailingSilence,
     rule3_min_utterance_length: rule3MinUtteranceLength,
     hotwords_file: toCPointer(hotwordsFile),
-    hotwords_score: hotwordsScore)
+    hotwords_score: hotwordsScore,
+    ctc_fst_decoder_config: ctcFstDecoderConfig
+  )
 }
 
 /// Wrapper for recognition result.

swift-api-examples/run-streaming-hlg-decode-file.sh

@@ -0,0 +1,36 @@
+#!/usr/bin/env bash
+
+set -ex
+
+if [ ! -d ../build-swift-macos ]; then
+  echo "Please run ../build-swift-macos.sh first!"
+  exit 1
+fi
+
+if [ ! -f ./sherpa-onnx-streaming-zipformer-ctc-small-2024-03-18/HLG.fst ]; then
+  echo "Downloading the pre-trained model for testing."
+
+  wget -q https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-streaming-zipformer-ctc-small-2024-03-18.tar.bz2
+  tar xvf sherpa-onnx-streaming-zipformer-ctc-small-2024-03-18.tar.bz2
+  rm sherpa-onnx-streaming-zipformer-ctc-small-2024-03-18.tar.bz2
+fi
+
+if [ ! -e ./streaming-hlg-decode-file ]; 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 \
+    ./streaming-hlg-decode-file.swift  ./SherpaOnnx.swift \
+    -L ../build-swift-macos/install/lib/ \
+    -l sherpa-onnx \
+    -l onnxruntime \
+    -o streaming-hlg-decode-file
+
+  strip ./streaming-hlg-decode-file
+else
+  echo "./streaming-hlg-decode-file exists - skip building"
+fi
+
+export DYLD_LIBRARY_PATH=$PWD/../build-swift-macos/install/lib:$DYLD_LIBRARY_PATH
+./streaming-hlg-decode-file

swift-api-examples/streaming-hlg-decode-file.swift

@@ -0,0 +1,79 @@
+import AVFoundation
+
+extension AudioBuffer {
+  func array() -> [Float] {
+    return Array(UnsafeBufferPointer(self))
+  }
+}
+
+extension AVAudioPCMBuffer {
+  func array() -> [Float] {
+    return self.audioBufferList.pointee.mBuffers.array()
+  }
+}
+
+func run() {
+  let filePath =
+    "./sherpa-onnx-streaming-zipformer-ctc-small-2024-03-18/test_wavs/8k.wav"
+  let model =
+    "./sherpa-onnx-streaming-zipformer-ctc-small-2024-03-18/ctc-epoch-30-avg-3-chunk-16-left-128.int8.onnx"
+  let tokens = "./sherpa-onnx-streaming-zipformer-ctc-small-2024-03-18/tokens.txt"
+  let zipfomer2CtcModelConfig = sherpaOnnxOnlineZipformer2CtcModelConfig(
+    model: model
+  )
+
+  let modelConfig = sherpaOnnxOnlineModelConfig(
+    tokens: tokens,
+    zipformer2Ctc: zipfomer2CtcModelConfig
+  )
+
+  let featConfig = sherpaOnnxFeatureConfig(
+    sampleRate: 16000,
+    featureDim: 80
+  )
+
+  let ctcFstDecoderConfig = sherpaOnnxOnlineCtcFstDecoderConfig(
+    graph: "./sherpa-onnx-streaming-zipformer-ctc-small-2024-03-18/HLG.fst",
+    maxActive: 3000
+  )
+
+  var config = sherpaOnnxOnlineRecognizerConfig(
+    featConfig: featConfig,
+    modelConfig: modelConfig,
+    ctcFstDecoderConfig: ctcFstDecoderConfig
+  )
+
+  let recognizer = SherpaOnnxRecognizer(config: &config)
+
+  let fileURL: NSURL = NSURL(fileURLWithPath: filePath)
+  let audioFile = try! AVAudioFile(forReading: fileURL as URL)
+
+  let audioFormat = audioFile.processingFormat
+  assert(audioFormat.channelCount == 1)
+  assert(audioFormat.commonFormat == AVAudioCommonFormat.pcmFormatFloat32)
+
+  let audioFrameCount = UInt32(audioFile.length)
+  let audioFileBuffer = AVAudioPCMBuffer(pcmFormat: audioFormat, frameCapacity: audioFrameCount)
+
+  try! audioFile.read(into: audioFileBuffer!)
+  let array: [Float]! = audioFileBuffer?.array()
+  recognizer.acceptWaveform(samples: array, sampleRate: Int(audioFormat.sampleRate))
+
+  let tailPadding = [Float](repeating: 0.0, count: 3200)
+  recognizer.acceptWaveform(samples: tailPadding, sampleRate: Int(audioFormat.sampleRate))
+
+  recognizer.inputFinished()
+  while recognizer.isReady() {
+    recognizer.decode()
+  }
+
+  let result = recognizer.getResult()
+  print("\nresult is:\n\(result.text)")
+}
+
+@main
+struct App {
+  static func main() {
+    run()
+  }
+}