Add emotion, event of SenseVoice. (#1257)

* Add emotion, event of SenseVoice.

* Fix tokens size check and update java api.

https://github.com/k2-fsa/sherpa-onnx/pull/1257

@@ -531,6 +531,20 @@ const SherpaOnnxOfflineRecognizerResult *SherpaOnnxGetOfflineStreamResult(
   c_lang[lang.size()] = '\0';
   r->lang = c_lang;
 
+  // emotion
+  const auto &emotion = result.emotion;
+  char *c_emotion = new char[emotion.size() + 1];
+  std::copy(emotion.begin(), emotion.end(), c_emotion);
+  c_emotion[emotion.size()] = '\0';
+  r->emotion = c_emotion;
+
+  // event
+  const auto &event = result.event;
+  char *c_event = new char[event.size() + 1];
+  std::copy(event.begin(), event.end(), c_event);
+  c_event[event.size()] = '\0';
+  r->event = c_event;
+
   // copy json
   std::string json = result.AsJsonString();
   char *pJson = new char[json.size() + 1];
@@ -588,6 +602,8 @@ void SherpaOnnxDestroyOfflineRecognizerResult(
     delete[] r->tokens_arr;
     delete[] r->json;
     delete[] r->lang;
+    delete[] r->emotion;
+    delete[] r->event;
     delete r;
   }
 }

@@ -544,6 +544,12 @@ SHERPA_ONNX_API typedef struct SherpaOnnxOfflineRecognizerResult {
 
   // return recognized language
   const char *lang;
+
+  // return emotion.
+  const char *emotion;
+
+  // return event.
+  const char *event;
 } SherpaOnnxOfflineRecognizerResult;
 
 /// Get the result of the offline stream.

@@ -52,6 +52,13 @@ static OfflineRecognitionResult ConvertSenseVoiceResult(
 
   r.words = std::move(src.words);
 
+  // parse lang, emotion and event from tokens.
+  if (src.tokens.size() >= 3) {
+    r.lang = sym_table[src.tokens[0]];
+    r.emotion = sym_table[src.tokens[1]];
+    r.event = sym_table[src.tokens[2]];
+  }
+
   return r;
 }
 

@@ -304,6 +304,19 @@ const OfflineRecognitionResult &OfflineStream::GetResult() const {
 std::string OfflineRecognitionResult::AsJsonString() const {
   std::ostringstream os;
   os << "{";
+
+  os << "\"lang\""
+     << ": ";
+  os << std::quoted(lang) << ", ";
+
+  os << "\"emotion\""
+     << ": ";
+  os << std::quoted(emotion) << ", ";
+
+  os << "\"event\""
+     << ": ";
+  os << std::quoted(event) << ", ";
+
   os << "\"text\""
      << ": ";
   os << std::quoted(text) << ", ";

@@ -28,6 +28,12 @@ struct OfflineRecognitionResult {
 
   std::string lang;
 
+  // emotion target of the audio.
+  std::string emotion;
+
+  // event target of the audio.
+  std::string event;
+
     /// timestamps.size() == tokens.size()
   /// timestamps[i] records the time in seconds when tokens[i] is decoded.
   std::vector<float> timestamps;

@@ -41,7 +41,10 @@ public class OfflineRecognizer {
         String text = (String) arr[0];
         String[] tokens = (String[]) arr[1];
         float[] timestamps = (float[]) arr[2];
-        return new OfflineRecognizerResult(text, tokens, timestamps);
+        String lang = (String) arr[3];
+        String emotion = (String) arr[4];
+        String event = (String) arr[5];
+        return new OfflineRecognizerResult(text, tokens, timestamps, lang, emotion, event);
     }
 
     private native void delete(long ptr);

@@ -6,11 +6,17 @@ public class OfflineRecognizerResult {
     private final String text;
     private final String[] tokens;
     private final float[] timestamps;
+    private final String lang;
+    private final String emotion;
+    private final String event;
 
-    public OfflineRecognizerResult(String text, String[] tokens, float[] timestamps) {
+    public OfflineRecognizerResult(String text, String[] tokens, float[] timestamps, String lang, String emotion, String event) {
         this.text = text;
         this.tokens = tokens;
         this.timestamps = timestamps;
+        this.lang = lang;
+        this.emotion = emotion;
+        this.event = event;
     }
 
     public String getText() {
@@ -24,4 +30,16 @@ public class OfflineRecognizerResult {
     public float[] getTimestamps() {
         return timestamps;
     }
+
+    public String getLang() {
+        return lang;
+    }
+
+    public String getEmotion() {
+        return emotion;
+    }
+
+    public String getEvent() {
+        return event;
+    }
 }

@@ -320,8 +320,11 @@ Java_com_k2fsa_sherpa_onnx_OfflineRecognizer_getResult(JNIEnv *env,
   // [0]: text, jstring
   // [1]: tokens, array of jstring
   // [2]: timestamps, array of float
+  // [3]: lang, jstring
+  // [4]: emotion, jstring
+  // [5]: event, jstring
   jobjectArray obj_arr = (jobjectArray)env->NewObjectArray(
-      3, env->FindClass("java/lang/Object"), nullptr);
+      6, env->FindClass("java/lang/Object"), nullptr);
 
   jstring text = env->NewStringUTF(result.text.c_str());
   env->SetObjectArrayElement(obj_arr, 0, text);
@@ -344,5 +347,12 @@ Java_com_k2fsa_sherpa_onnx_OfflineRecognizer_getResult(JNIEnv *env,
 
   env->SetObjectArrayElement(obj_arr, 2, timestamps_arr);
 
+  // [3]: lang, jstring
+  // [4]: emotion, jstring
+  // [5]: event, jstring
+  env->SetObjectArrayElement(obj_arr, 3, env->NewStringUTF(result.lang.c_str()));
+  env->SetObjectArrayElement(obj_arr, 4, env->NewStringUTF(result.emotion.c_str()));
+  env->SetObjectArrayElement(obj_arr, 5, env->NewStringUTF(result.event.c_str()));
+
   return obj_arr;
 }

@@ -6,6 +6,9 @@ data class OfflineRecognizerResult(
     val text: String,
     val tokens: Array<String>,
     val timestamps: FloatArray,
+    val lang: String,
+    val emotion: String,
+    val event: String,
 )
 
 data class OfflineTransducerModelConfig(
@@ -96,7 +99,10 @@ class OfflineRecognizer(
         val text = objArray[0] as String
         val tokens = objArray[1] as Array<String>
         val timestamps = objArray[2] as FloatArray
-        return OfflineRecognizerResult(text = text, tokens = tokens, timestamps = timestamps)
+        val lang = objArray[3] as String
+        val emotion = objArray[4] as String
+        val event = objArray[5] as String
+        return OfflineRecognizerResult(text = text, tokens = tokens, timestamps = timestamps, lang = lang, emotion = emotion, event = event)
     }
 
     fun decode(stream: OfflineStream) = decode(ptr, stream.ptr)

@@ -32,6 +32,12 @@ static void PybindOfflineRecognitionResult(py::module *m) {  // NOLINT
             return py::str(PyUnicode_DecodeUTF8(self.text.c_str(),
                                                 self.text.size(), "ignore"));
           })
+      .def_property_readonly("lang",
+                            [](const PyClass &self) { return self.lang; })
+      .def_property_readonly("emotion",
+                            [](const PyClass &self) { return self.emotion; })
+      .def_property_readonly("event",
+                            [](const PyClass &self) { return self.event; })
       .def_property_readonly("tokens",
                              [](const PyClass &self) { return self.tokens; })
       .def_property_readonly("words",