Android용 오디오 분류 가이드

AudioClassifierOptions options =
    AudioClassifierOptions.builder()
        .setBaseOptions(
            BaseOptions.builder().setModelAssetPath("model.tflite").build())
        .setRunningMode(RunningMode.AUDIO_CLIPS)
        .setMaxResults(5)
        .build();
audioClassifier = AudioClassifier.createFromOptions(context, options);
    

AudioClassifierOptions options =
    AudioClassifierOptions.builder()
        .setBaseOptions(
            BaseOptions.builder().setModelAssetPath("model.tflite").build())
        .setRunningMode(RunningMode.AUDIO_STREAM)
        .setMaxResults(5)
        .setResultListener(audioClassifierResult -> {
             // Process the classification result here.
        })
        .build();
audioClassifier = AudioClassifier.createFromOptions(context, options);
    

import com.google.mediapipe.tasks.components.containers.AudioData;

// Load an audio on the user’s device as a float array.

// Convert a float array to a MediaPipe’s AudioData object.
AudioData audioData =
    AudioData.create(
        AudioData.AudioDataFormat.builder()
            .setNumOfChannels(numOfChannels)
            .setSampleRate(sampleRate)
            .build(),
        floatData.length);
audioData.load(floatData);
    

import android.media.AudioRecord;
import com.google.mediapipe.tasks.components.containers.AudioData;

AudioRecord audioRecord =
    audioClassifier.createAudioRecord(/* numChannels= */ 1, /* sampleRate= */ 16000);
audioRecord.startRecording();

...

// To get a one second clip from the AudioRecord object:
AudioData audioData =
    AudioData.create(
        16000 /*sample counts per second*/);
        AudioData.AudioDataFormat.create(audioRecord.getFormat()),
audioData.load(audioRecord)
    

AudioClassifierResult classifierResult = audioClassifier.classify(audioData);
    

// Run inference on the audio block. The classifications results will be available
// via the `resultListener` provided in the `AudioClassifierOptions` when
// the audio classifier was created.
audioClassifier.classifyAsync(audioBlock, timestampMs);
    

// In the audio clips mode, the classification results are for the entire audio
// clip. The results are timestamped AudioClassifierResult objects, each
// classifying an interval of the entire audio clip that starts at
// ClassificationResult.timestampMs().get().

for (ClassificationResult result : audioClassifierResult.classificationResults()) {
  // Audio interval start timestamp:
  result.timestampMs().get();
  // Classification result of the audio interval.
  result.classifications();
}
    

// In the audio stream mode, the classification results list only contains one
// element, representing the classification result of the audio block that
// starts at ClassificationResult.timestampMs in the audio stream.

ClassificationResult result = audioClassifierResult.classificationResults().get(0);
// The audio block start timestamp
audioClassifierResult.timestampMs();
// Alternatively, the same timestamp can be retrieved from
// result.timestampMs().get();

// Classification result.
result.classifications();