执行 alizeSystem.createSpeakerModel(userName) 时因 "UBM not found" 而崩溃；

Question

我正在使用 android-alize 作为说话人识别应用。正如文档所说，我正在尝试训练如下模型，

// 向系统发送音频 alizeSystem.addAudio（音频）；

// 用音频训练模型 alizeSystem.createSpeakerModel("Somebody");

但是我的应用程序因以下堆栈跟踪而崩溃，

04-17 14:29:31.216 5546-5546/com.example.channa.poc_alize W/IInputConnectionWrapper: getTextAfterCursor on inactive InputConnection
04-17 14:29:31.253 5546-5546/com.example.channa.poc_alize W/IInputConnectionWrapper: reportFullscreenMode on inexistent InputConnection
04-17 14:29:31.253 5546-5546/com.example.channa.poc_alize W/IInputConnectionWrapper: finishComposingText on inactive InputConnection
04-17 14:29:33.184 5546-5546/com.example.channa.poc_alize I/System.out: content://com.android.externalstorage.documents/document/primary%3AAudioClipsForSpeakerRecognition%2F1%2Fspeaker_1.wav
04-17 14:29:34.155 5546-5546/com.example.channa.poc_alize W/System.err: AlizeSpkRec.AlizeException: [ Exception 0x7c4ca25480 ]
      message   = "UBM not found"
      source file = /Users/channa/Documents/Projects/Work/Sample Projects/android-alize/alize/src/main/cpp/LIA_RAL/LIA_SpkDet/SimpleSpkDetSystem/src/SimpleSpkDetSystem.cpp
      line number = 869
        at AlizeSpkRec.SimpleSpkDetSystem.createSpeakerModel(Native Method)
        at com.example.channa.poc_alize.SpeakerRecognition.trainSpeakerModel(SpeakerRecognition.java:40)
        at com.example.channa.poc_alize.MainActivity.onActivityResult(MainActivity.java:97)
        at android.app.Activity.dispatchActivityResult(Activity.java:7226)
        at android.app.ActivityThread.deliverResults(ActivityThread.java:4477)
        at android.app.ActivityThread.handleSendResult(ActivityThread.java:4524)
        at android.app.ActivityThread.-wrap22(ActivityThread.java)
        at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1691)
        at android.os.Handler.dispatchMessage(Handler.java:102)
        at android.os.Looper.loop(Looper.java:154)
        at android.app.ActivityThread.main(ActivityThread.java:6692)
        at java.lang.reflect.Method.invoke(Native Method)
        at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:1468)
        at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:1358)
04-17 14:29:34.182 5546-5586/com.example.channa.poc_alize D/mali_winsys: EGLint new_window_surface(egl_winsys_display*, void*, EGLSurface, EGLConfig, egl_winsys_surface**, egl_color_buffer_format*, EGLBoolean) returns 0x3000,  [1440x2560]-format:1
04-17 14:29:34.183 5546-5546/com.example.channa.poc_alize D/ViewRootImpl@52672bd[MainActivity]: MSG_RESIZED_REPORT: ci=Rect(0, 96 - 0, 0) vi=Rect(0, 96 - 0, 0) or=1
    MSG_WINDOW_FOCUS_CHANGED 1

感谢您的意见

Answer 1

进一步调试Alize-LIA_RAL，发现android项目在assets/gmm/world.gmm中寻找world.gmm（UBM）。似乎我们必须使用 LIA_RAL 库生成此 world.gmm，然后将其添加到 android 代码中。

但是我们必须在 alize 初始化期间使用下面给定的代码片段指定 UBM 位置。

    InputStream backgroundModelAsset = null;
    try {
        backgroundModelAsset = context.getAssets().open("gmm/world.gmm");
        alizeSystem.loadBackgroundModel(backgroundModelAsset);
        backgroundModelAsset.close();
    } catch (IOException e) {
        e.printStackTrace();
    } catch (AlizeException e) {
        e.printStackTrace();
    }

我设法从 google 搜索中找到了一个 world.gmm，当我将它添加到我的 android 项目时，错误消失了。然而，预测并不准确。

因此，我正在进一步研究 LIA_RAL 源代码以通过提供满足我要求的训练数据（音频样本）来生成此 world.gmm (UBM)。

执行 alizeSystem.createSpeakerModel(userName) 时因 "UBM not found" 而崩溃；

Crashing with "UBM not found" when executing alizeSystem.createSpeakerModel(userName);

android

speech-recognition

machine-learning