Update C API to include timestamps for offline ASR

sherpa-onnx/c-api/c-api.cc

@@ -353,11 +353,22 @@ SherpaOnnxOfflineRecognizerResult *GetOfflineStreamResult(
   std::copy(text.begin(), text.end(), const_cast<char *>(r->text));
   const_cast<char *>(r->text)[text.size()] = 0;
 
+  if (!result.timestamps.empty()) {
+    r->timestamps = new float[result.timestamps.size()];
+    std::copy(result.timestamps.begin(), result.timestamps.end(),
+              r->timestamps);
+    r->count = result.timestamps.size();
+  } else {
+    r->timestamps = nullptr;
+    r->count = 0;
+  }
+
   return r;
 }
 
 void DestroyOfflineRecognizerResult(
     const SherpaOnnxOfflineRecognizerResult *r) {
   delete[] r->text;
+  delete[] r->timestamps;
   delete r;
 }

sherpa-onnx/c-api/c-api.h

@@ -408,6 +408,14 @@ SHERPA_ONNX_API void DecodeMultipleOfflineStreams(
 
 SHERPA_ONNX_API typedef struct SherpaOnnxOfflineRecognizerResult {
   const char *text;
+
+  // Pointer to continuous memory which holds timestamps
+  //
+  // It is NULL if the model does not support timestamps
+  float *timestamps;
+
+  // number of entries in timestamps
+  int32_t count;
   // TODO(fangjun): Add more fields
 } SherpaOnnxOfflineRecognizerResult;
 

sherpa-onnx/csrc/offline-paraformer-greedy-search-decoder.cc

@@ -5,6 +5,7 @@
 #include "sherpa-onnx/csrc/offline-paraformer-greedy-search-decoder.h"
 
 #include <algorithm>
+#include <utility>
 #include <vector>
 
 #include "sherpa-onnx/csrc/macros.h"

sherpa-onnx/csrc/offline-paraformer-model.cc

@@ -6,6 +6,7 @@
 
 #include <algorithm>
 #include <string>
+#include <utility>
 
 #include "sherpa-onnx/csrc/macros.h"
 #include "sherpa-onnx/csrc/onnx-utils.h"

swift-api-examples/SherpaOnnx.swift

@@ -349,6 +349,23 @@ class SherpaOnnxOfflineRecongitionResult {
     return String(cString: result.pointee.text)
   }
 
+  var count: Int32 {
+    return result.pointee.count
+  }
+
+  var timestamps: [Float] {
+    if let p = result.pointee.timestamps {
+      var timestamps: [Float] = []
+      for index in 0..<count {
+        timestamps.append(p[Int(index)])
+      }
+      return timestamps
+    } else {
+      let timestamps: [Float] = []
+      return timestamps
+    }
+  }
+
   init(result: UnsafePointer<SherpaOnnxOfflineRecognizerResult>!) {
     self.result = result
   }

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

@@ -13,21 +13,45 @@ extension AVAudioPCMBuffer {
 }
 
 func run() {
-  let encoder = "./sherpa-onnx-whisper-tiny.en/tiny.en-encoder.int8.onnx"
-  let decoder = "./sherpa-onnx-whisper-tiny.en/tiny.en-decoder.int8.onnx"
-  let tokens = "./sherpa-onnx-whisper-tiny.en/tiny.en-tokens.txt"
 
-  let whisperConfig = sherpaOnnxOfflineWhisperModelConfig(
-    encoder: encoder,
-    decoder: decoder
-  )
+  var recognizer: SherpaOnnxOfflineRecognizer
+  var modelConfig: SherpaOnnxOfflineModelConfig
+  var modelType = "whisper"
+  // modelType = "paraformer"
 
-  let modelConfig = sherpaOnnxOfflineModelConfig(
-    tokens: tokens,
-    whisper: whisperConfig,
-    debug: 0,
-    modelType: "whisper"
-  )
+  if modelType == "whisper" {
+    let encoder = "./sherpa-onnx-whisper-tiny.en/tiny.en-encoder.int8.onnx"
+    let decoder = "./sherpa-onnx-whisper-tiny.en/tiny.en-decoder.int8.onnx"
+    let tokens = "./sherpa-onnx-whisper-tiny.en/tiny.en-tokens.txt"
+
+    let whisperConfig = sherpaOnnxOfflineWhisperModelConfig(
+      encoder: encoder,
+      decoder: decoder
+    )
+
+    modelConfig = sherpaOnnxOfflineModelConfig(
+      tokens: tokens,
+      whisper: whisperConfig,
+      debug: 0,
+      modelType: "whisper"
+    )
+  } else if modelType == "paraformer" {
+    let model = "./sherpa-onnx-paraformer-zh-2023-09-14/model.int8.onnx"
+    let tokens = "./sherpa-onnx-paraformer-zh-2023-09-14/tokens.txt"
+    let paraformerConfig = sherpaOnnxOfflineParaformerModelConfig(
+      model: model
+    )
+
+    modelConfig = sherpaOnnxOfflineModelConfig(
+      tokens: tokens,
+      paraformer: paraformerConfig,
+      debug: 0,
+      modelType: "paraformer"
+    )
+  } else {
+    print("Please specify a supported modelType \(modelType)")
+    return
+  }
 
   let featConfig = sherpaOnnxFeatureConfig(
     sampleRate: 16000,
@@ -38,7 +62,7 @@ func run() {
     modelConfig: modelConfig
   )
 
-  let recognizer = SherpaOnnxOfflineRecognizer(config: &config)
+  recognizer = SherpaOnnxOfflineRecognizer(config: &config)
 
   let filePath = "./sherpa-onnx-whisper-tiny.en/test_wavs/0.wav"
   let fileURL: NSURL = NSURL(fileURLWithPath: filePath)
@@ -55,6 +79,10 @@ func run() {
   let array: [Float]! = audioFileBuffer?.array()
   let result = recognizer.decode(samples: array, sampleRate: Int(audioFormat.sampleRate))
   print("\nresult is:\n\(result.text)")
+  if result.timestamps.count != 0 {
+    print("\ntimestamps is:\n\(result.timestamps)")
+  }
+
 }
 
 @main