whisper.rn 0.1.5 → 0.2.0

package/README.md

@@ -20,6 +20,23 @@ npm install whisper.rn
 
 Then re-run `npx pod-install` again for iOS.
 
+## Add Microphone Permissions (Optional)
+
+If you want to use realtime transcribe, you need to add the microphone permission to your app.
+
+### iOS
+Add these lines to ```ios/[YOU_APP_NAME]/info.plist```
+```xml
+<key>NSMicrophoneUsageDescription</key>
+<string>This app requires microphone access in order to transcribe speech</string>
+```
+
+### Android
+Add the following line to ```android/app/src/main/AndroidManifest.xml```
+```xml
+<uses-permission android:name="android.permission.RECORD_AUDIO" />
+```
+
 ## Usage
 
 ```js
@@ -30,13 +47,35 @@ const sampleFilePath = 'file://.../sample.wav'
 
 const whisperContext = await initWhisper({ filePath })
 
-const { result } = await whisperContext.transcribe(sampleFilePath, {
-  language: 'en',
-  // More options
-})
+const options = { language: 'en' }
+const { stop, promise } = whisperContext.transcribe(sampleFilePath, options)
+
+const { result } = await promise
 // result: (The inference text result from audio file)
 ```
 
+Use realtime transcribe:
+
+```js
+const { stop, subscribe } = whisperContext.transcribeRealtime(options)
+
+subscribe(evt => {
+  const { isCapturing, data, processTime, recordingTime } = evt
+  console.log(
+    `Realtime transcribing: ${isCapturing ? 'ON' : 'OFF'}\n` +
+      // The inference text result from audio record:
+      `Result: ${data.result}\n\n` + 
+      `Process time: ${processTime}ms\n` +
+      `Recording time: ${recordingTime}ms`,
+  )
+  if (!isCapturing) console.log('Finished realtime transcribing')
+})
+```
+
+In Android, you may need to request the microphone permission by [`PermissionAndroid`](https://reactnative.dev/docs/permissionsandroid).
+
+The documentation is not ready yet, please see the comments of [index](./src/index.tsx) file for more details at the moment.
+
 ## Run with example
 
 The example app is using [react-native-fs](https://github.com/itinance/react-native-fs) to download the model file and audio file.

package/android/build.gradle

@@ -40,10 +40,8 @@ android {
     buildConfigField "boolean", "IS_NEW_ARCHITECTURE_ENABLED", isNewArchitectureEnabled().toString()
   }
   externalNativeBuild {
-    externalNativeBuild {
-      ndkBuild {
-        path 'src/main/jni/whisper/Android.mk'
-      }
+    ndkBuild {
+      path 'src/main/jni/whisper/Android.mk'
     }
   }
   buildTypes {

package/android/src/main/java/com/rnwhisper/RNWhisperModule.java

@@ -5,6 +5,7 @@ import android.util.Log;
 import android.os.Build;
 import android.os.Handler;
 import android.os.AsyncTask;
+import android.media.AudioRecord;
 
 import com.facebook.react.bridge.Promise;
 import com.facebook.react.bridge.ReactApplicationContext;
@@ -51,7 +52,7 @@ public class RNWhisperModule extends ReactContextBaseJavaModule implements Lifec
             throw new Exception("Failed to initialize context");
           }
           int id = Math.abs(new Random().nextInt());
-          WhisperContext whisperContext = new WhisperContext(context);
+          WhisperContext whisperContext = new WhisperContext(id, reactContext, context);
           contexts.put(id, whisperContext);
           return id;
         } catch (Exception e) {
@@ -72,18 +73,27 @@ public class RNWhisperModule extends ReactContextBaseJavaModule implements Lifec
   }
 
   @ReactMethod
-  public void transcribe(int id, String filePath, ReadableMap options, Promise promise) {
+  public void transcribeFile(int id, int jobId, String filePath, ReadableMap options, Promise promise) {
+    final WhisperContext context = contexts.get(id);
+    if (context == null) {
+      promise.reject("Context not found");
+      return;
+    }
+    if (context.isCapturing()) {
+      promise.reject("The context is in realtime transcribe mode");
+      return;
+    }
+    if (context.isTranscribing()) {
+      promise.reject("Context is already transcribing");
+      return;
+    }
     new AsyncTask<Void, Void, WritableMap>() {
       private Exception exception;
 
       @Override
       protected WritableMap doInBackground(Void... voids) {
         try {
-          WhisperContext context = contexts.get(id);
-          if (context == null) {
-            throw new Exception("Context " + id + " not found");
-          }
-          return context.transcribe(filePath, options);
+          return context.transcribeFile(jobId, filePath, options);
         } catch (Exception e) {
           exception = e;
           return null;
@@ -101,6 +111,35 @@ public class RNWhisperModule extends ReactContextBaseJavaModule implements Lifec
     }.execute();
   }
 
+  @ReactMethod
+  public void startRealtimeTranscribe(int id, int jobId, ReadableMap options, Promise promise) {
+    final WhisperContext context = contexts.get(id);
+    if (context == null) {
+      promise.reject("Context not found");
+      return;
+    }
+    if (context.isCapturing()) {
+      promise.reject("Context is already in capturing");
+      return;
+    }
+    int state = context.startRealtimeTranscribe(jobId, options);
+    if (state == AudioRecord.STATE_INITIALIZED) {
+      promise.resolve(null);
+      return;
+    }
+    promise.reject("Failed to start realtime transcribe. State: " + state);
+  }
+
+  @ReactMethod
+  public void abortTranscribe(int contextId, int jobId, Promise promise) {
+    WhisperContext context = contexts.get(contextId);
+    if (context == null) {
+      promise.reject("Context not found");
+      return;
+    }
+    context.stopTranscribe(jobId);
+  }
+
   @ReactMethod
   public void releaseContext(int id, Promise promise) {
     new AsyncTask<Void, Void, Void>() {
@@ -168,6 +207,7 @@ public class RNWhisperModule extends ReactContextBaseJavaModule implements Lifec
 
   @Override
   public void onHostDestroy() {
+    WhisperContext.abortAllTranscribe();
     for (WhisperContext context : contexts.values()) {
       context.release();
     }

package/android/src/main/java/com/rnwhisper/WhisperContext.java

@@ -4,10 +4,15 @@ import com.facebook.react.bridge.Arguments;
 import com.facebook.react.bridge.WritableArray;
 import com.facebook.react.bridge.WritableMap;
 import com.facebook.react.bridge.ReadableMap;
+import com.facebook.react.bridge.ReactApplicationContext;
+import com.facebook.react.modules.core.DeviceEventManagerModule;
 
 import android.util.Log;
 import android.os.Build;
 import android.content.res.AssetManager;
+import android.media.AudioFormat;
+import android.media.AudioRecord;
+import android.media.MediaRecorder.AudioSource;
 
 import java.util.Random;
 import java.lang.StringBuilder;
@@ -26,16 +31,175 @@ import java.nio.ShortBuffer;
 
 public class WhisperContext {
   public static final String NAME = "RNWhisperContext";
+
+  private static final int SAMPLE_RATE = 16000;
+  private static final int CHANNEL_CONFIG = AudioFormat.CHANNEL_IN_MONO;
+  private static final int AUDIO_FORMAT = AudioFormat.ENCODING_PCM_16BIT;
+  private static final int AUDIO_SOURCE = AudioSource.VOICE_RECOGNITION;
+  private static final int DEFAULT_MAX_AUDIO_SEC = 30;
+
+  private int id;
+  private ReactApplicationContext reactContext;
   private long context;
 
-  public WhisperContext(long context) {
+  private DeviceEventManagerModule.RCTDeviceEventEmitter eventEmitter;
+  
+  private int jobId = -1;
+  private AudioRecord recorder = null;
+  private int bufferSize;
+  private short[] buffer16;
+  private int nSamples = 0;
+  private boolean isCapturing = false;
+  private boolean isTranscribing = false;
+  private boolean isRealtime = false;
+
+  public WhisperContext(int id, ReactApplicationContext reactContext, long context) {
+    this.id = id;
     this.context = context;
+    this.reactContext = reactContext;
+    eventEmitter = reactContext.getJSModule(DeviceEventManagerModule.RCTDeviceEventEmitter.class);
+    bufferSize = AudioRecord.getMinBufferSize(SAMPLE_RATE, CHANNEL_CONFIG, AUDIO_FORMAT);
+  }
+
+  public int startRealtimeTranscribe(int jobId, ReadableMap options) {
+    if (isCapturing || isTranscribing) {
+      return -100;
+    }
+
+    recorder = new AudioRecord(AUDIO_SOURCE, SAMPLE_RATE, CHANNEL_CONFIG, AUDIO_FORMAT, bufferSize);
+
+    int state = recorder.getState();
+    if (state != AudioRecord.STATE_INITIALIZED) {
+      recorder.release();
+      return state;
+    }
+    
+    int realtimeAudioSec = options.hasKey("realtimeAudioSec") ? options.getInt("realtimeAudioSec") : 0;
+    final int maxAudioSec = realtimeAudioSec > 0 ? realtimeAudioSec : DEFAULT_MAX_AUDIO_SEC;
+
+    buffer16 = new short[maxAudioSec * SAMPLE_RATE * Short.BYTES];
+
+    this.jobId = jobId;
+    isCapturing = true;
+    isRealtime = true;
+    nSamples = 0;
+  
+    recorder.startRecording();
+
+    new Thread(new Runnable() {
+      @Override
+      public void run() {
+        try {
+          short[] buffer = new short[bufferSize];
+          Thread fullHandler = null;
+          while (isCapturing) {
+            try {
+              int n = recorder.read(buffer, 0, bufferSize);
+              if (n == 0) continue;
+
+              if (nSamples + n > maxAudioSec * SAMPLE_RATE) {
+                // Full, ignore data
+                isCapturing = false;
+                if (!isTranscribing)
+                  emitTranscribeEvent("@RNWhisper_onRealtimeTranscribeEnd", Arguments.createMap());
+                break;
+              }
+              nSamples += n;
+              for (int i = 0; i < n; i++) {
+                buffer16[nSamples + i] = buffer[i];
+              }
+              if (!isTranscribing && nSamples > SAMPLE_RATE / 2) {
+                isTranscribing = true;
+                Log.d(NAME, "Start transcribing realtime: " + nSamples);
+                fullHandler = new Thread(new Runnable() {
+                  @Override
+                  public void run() {
+                    if (!isCapturing) return;
+
+                    // convert I16 to F32
+                    float[] nSamplesBuffer32 = new float[nSamples];
+                    for (int i = 0; i < nSamples; i++) {
+                      nSamplesBuffer32[i] = buffer16[i] / 32768.0f;
+                    }
+
+                    int timeStart = (int) System.currentTimeMillis();
+                    int code = full(jobId, options, nSamplesBuffer32, nSamples);
+                    int timeEnd = (int) System.currentTimeMillis();
+                    int timeRecording = (int) (nSamples / SAMPLE_RATE * 1000);
+
+                    WritableMap payload = Arguments.createMap();
+                    payload.putBoolean("isCapturing", isCapturing);
+                    payload.putInt("code", code);
+                    payload.putInt("processTime", timeEnd - timeStart);
+                    payload.putInt("recordingTime", timeRecording);
+
+                    if (code == 0) {
+                      payload.putMap("data", getTextSegments());
+                      emitTranscribeEvent("@RNWhisper_onRealtimeTranscribe", payload);
+                    } else {
+                      payload.putString("error", "Transcribe failed with code " + code);
+                      emitTranscribeEvent("@RNWhisper_onRealtimeTranscribe", payload);
+                    }
+
+                    if (!isCapturing) {
+                      emitTranscribeEvent("@RNWhisper_onRealtimeTranscribeEnd", Arguments.createMap());
+                    }
+                    isTranscribing = false;
+                  }
+                });
+                fullHandler.start();
+              }
+            } catch (Exception e) {
+              Log.e(NAME, "Error transcribing realtime: " + e.getMessage());
+            }
+          }
+          if (fullHandler != null) {
+            fullHandler.join(); // Wait for full transcribe to finish
+          }
+          recorder.stop();
+        } catch (Exception e) {
+          e.printStackTrace();
+        } finally {
+          recorder.release();
+          recorder = null;
+        }
+      }
+    }).start();
+
+    return state;
+  }
+
+  private void emitTranscribeEvent(final String eventName, final WritableMap payload) {
+    WritableMap event = Arguments.createMap();
+    event.putInt("contextId", WhisperContext.this.id);
+    event.putInt("jobId", jobId);
+    event.putMap("payload", payload);
+    eventEmitter.emit(eventName, event);
   }
 
-  public WritableMap transcribe(final String filePath, final ReadableMap options) throws IOException, Exception {
-    int code = fullTranscribe(
+  public WritableMap transcribeFile(int jobId, String filePath, ReadableMap options) throws IOException, Exception {
+    this.jobId = jobId;
+    isTranscribing = true;
+    float[] audioData = decodeWaveFile(new File(filePath));
+    int code = full(jobId, options, audioData, audioData.length);
+    isTranscribing = false;
+    this.jobId = -1;
+    if (code != 0) {
+      throw new Exception("Failed to transcribe the file. Code: " + code);
+    }
+    return getTextSegments();
+  }
+
+  private int full(int jobId, ReadableMap options, float[] audioData, int audioDataLen) {
+    return fullTranscribe(
+      jobId,
       context,
-      decodeWaveFile(new File(filePath)),
+      // jboolean realtime,
+      isRealtime,
+      // float[] audio_data,
+      audioData,
+      // jint audio_data_len,
+      audioDataLen,
       // jint n_threads,
       options.hasKey("maxThreads") ? options.getInt("maxThreads") : -1,
       // jint max_context,
@@ -69,9 +233,9 @@ public class WhisperContext {
       // jstring prompt
       options.hasKey("prompt") ? options.getString("prompt") : null
     );
-    if (code != 0) {
-      throw new Exception("Transcription failed with code " + code);
-    }
+  }
+
+  private WritableMap getTextSegments() {
     Integer count = getTextSegmentCount(context);
     StringBuilder builder = new StringBuilder();
 
@@ -92,7 +256,27 @@ public class WhisperContext {
     return data;
   }
 
+
+  public boolean isCapturing() {
+    return isCapturing;
+  }
+
+  public boolean isTranscribing() {
+    return isTranscribing;
+  }
+
+  public void stopTranscribe(int jobId) {
+    abortTranscribe(jobId);
+    isCapturing = false;
+    isTranscribing = false;
+  }
+
+  public void stopCurrentTranscribe() {
+    stopTranscribe(this.jobId);
+  }
+
   public void release() {
+    stopCurrentTranscribe();
     freeContext(context);
   }
 
@@ -185,8 +369,11 @@ public class WhisperContext {
 
   protected static native long initContext(String modelPath);
   protected static native int fullTranscribe(
+    int job_id,
     long context,
+    boolean realtime,
     float[] audio_data,
+    int audio_data_len,
     int n_threads,
     int max_context,
     int word_thold,
@@ -203,6 +390,8 @@ public class WhisperContext {
     String language,
     String prompt
   );
+  protected static native void abortTranscribe(int jobId);
+  protected static native void abortAllTranscribe();
   protected static native int getTextSegmentCount(long context);
   protected static native String getTextSegment(long context, int index);
   protected static native int getTextSegmentT0(long context, int index);

package/android/src/main/jni/whisper/Whisper.mk

@@ -12,7 +12,7 @@ ifneq ($(APP_OPTIM),debug)
 endif
 
 LOCAL_CFLAGS    += -DSTDC_HEADERS -std=c11 -I $(WHISPER_LIB_DIR)
-LOCAL_CPPFLAGS  += -std=c++11
+LOCAL_CPPFLAGS  += -std=c++11 -I $(WHISPER_LIB_DIR)
 LOCAL_SRC_FILES := $(WHISPER_LIB_DIR)/ggml.c \
                    $(WHISPER_LIB_DIR)/whisper.cpp \
                    $(WHISPER_LIB_DIR)/rn-whisper.cpp \

package/android/src/main/jni/whisper/jni.cpp

@@ -36,8 +36,11 @@ JNIEXPORT jint JNICALL
 Java_com_rnwhisper_WhisperContext_fullTranscribe(
     JNIEnv *env,
     jobject thiz,
+    jint job_id,
     jlong context_ptr,
+    jboolean realtime,
     jfloatArray audio_data,
+    jint audio_data_len,
     jint n_threads,
     jint max_context,
     int word_thold,
@@ -57,7 +60,6 @@ Java_com_rnwhisper_WhisperContext_fullTranscribe(
     UNUSED(thiz);
     struct whisper_context *context = reinterpret_cast<struct whisper_context *>(context_ptr);
     jfloat *audio_data_arr = env->GetFloatArrayElements(audio_data, nullptr);
-    const jsize audio_data_length = env->GetArrayLength(audio_data);
 
     int max_threads = min(4, get_nprocs());
 
@@ -81,7 +83,7 @@ Java_com_rnwhisper_WhisperContext_fullTranscribe(
     params.speed_up = speed_up;
     params.offset_ms = 0;
     params.no_context = true;
-    params.single_segment = false;
+    params.single_segment = realtime;
 
     if (max_len > -1) {
         params.max_len = max_len;
@@ -117,19 +119,45 @@ Java_com_rnwhisper_WhisperContext_fullTranscribe(
         );
     }
 
+    params.encoder_begin_callback = [](struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, void * user_data) {
+        bool is_aborted = *(bool*)user_data;
+        return !is_aborted;
+    };
+    params.encoder_begin_callback_user_data = rn_whisper_assign_abort_map(job_id);
+
     LOGI("About to reset timings");
     whisper_reset_timings(context);
 
     LOGI("About to run whisper_full");
-    int code = whisper_full(context, params, audio_data_arr, audio_data_length);
+    int code = whisper_full(context, params, audio_data_arr, audio_data_len);
     if (code == 0) {
         // whisper_print_timings(context);
     }
     env->ReleaseFloatArrayElements(audio_data, audio_data_arr, JNI_ABORT);
     env->ReleaseStringUTFChars(language, language_chars);
+    rn_whisper_remove_abort_map(job_id);
     return code;
 }
 
+JNIEXPORT void JNICALL
+Java_com_rnwhisper_WhisperContext_abortTranscribe(
+    JNIEnv *env,
+    jobject thiz,
+    jint job_id
+) {
+    UNUSED(thiz);
+    rn_whisper_abort_transcribe(job_id);
+}
+
+JNIEXPORT void JNICALL
+Java_com_rnwhisper_WhisperContext_abortAllTranscribe(
+    JNIEnv *env,
+    jobject thiz
+) {
+    UNUSED(thiz);
+    rn_whisper_abort_all_transcribe();
+}
+
 JNIEXPORT jint JNICALL
 Java_com_rnwhisper_WhisperContext_getTextSegmentCount(
         JNIEnv *env, jobject thiz, jlong context_ptr) {
@@ -176,4 +204,4 @@ Java_com_rnwhisper_WhisperContext_freeContext(
     whisper_free(context);
 }
 
-} // extern "C"
+} // extern "C"

package/cpp/rn-whisper.cpp

@@ -1,6 +1,7 @@
 #include <cstdio>
 #include <string>
 #include <vector>
+#include <unordered_map>
 #include "whisper.h"
 
 extern "C" {
@@ -28,4 +29,29 @@ void rn_whisper_convert_prompt(
   }
 }
 
+std::unordered_map<int, bool> abort_map;
+
+bool* rn_whisper_assign_abort_map(int job_id) {
+  abort_map[job_id] = false;
+  return &abort_map[job_id];
+}
+
+void rn_whisper_remove_abort_map(int job_id) {
+  if (abort_map.find(job_id) != abort_map.end()) {
+    abort_map.erase(job_id);
+  }
+}
+
+void rn_whisper_abort_transcribe(int job_id) {
+  if (abort_map.find(job_id) != abort_map.end()) {
+    abort_map[job_id] = true;
+  }
+}
+
+void rn_whisper_abort_all_transcribe() {
+  for (auto it = abort_map.begin(); it != abort_map.end(); ++it) {
+    it->second = true;
+  }
+}
+
 }

package/cpp/rn-whisper.h

@@ -11,6 +11,11 @@ void rn_whisper_convert_prompt(
   std::string * prompt
 );
 
+bool* rn_whisper_assign_abort_map(int job_id);
+void rn_whisper_remove_abort_map(int job_id);
+void rn_whisper_abort_transcribe(int job_id);
+void rn_whisper_abort_all_transcribe();
+
 #ifdef __cplusplus
 }
 #endif

package/ios/RNWhisper.h

@@ -3,9 +3,9 @@
 #import "rn-whisper.h"
 #endif
 
-
 #import <React/RCTBridgeModule.h>
+#import <React/RCTEventEmitter.h>
 
-@interface RNWhisper : NSObject <RCTBridgeModule>
+@interface RNWhisper : RCTEventEmitter <RCTBridgeModule>
 
 @end