Python vggish_slim.define_vggish_slim() Examples

def __init__(self, checkpoint, pca_params, input_tensor_name, output_tensor_name):
        """Create a new Graph and a new Session for every VGGishExtractor object."""
        super(VGGishExtractor, self).__init__()
        
        self.graph = tf.Graph()
        with self.graph.as_default():
            vggish_slim.define_vggish_slim(training=False)

        sess_config = tf.ConfigProto(allow_soft_placement=True)
        sess_config.gpu_options.allow_growth = True
        self.sess = tf.Session(graph=self.graph, config=sess_config)
        vggish_slim.load_defined_vggish_slim_checkpoint(self.sess, checkpoint)
        
        # use the self.sess to init others
        self.input_tensor = self.graph.get_tensor_by_name(input_tensor_name)
        self.output_tensor = self.graph.get_tensor_by_name(output_tensor_name)

        # postprocessor
        self.postprocess = vggish_postprocess.Postprocessor(pca_params) 

def extract_vggish_features(paths, path2gt, model): 
    """Extracts VGGish features and their corresponding ground_truth and identifiers (the path).

       VGGish features are extracted from non-overlapping audio patches of 0.96 seconds, 
       where each audio patch covers 64 mel bands and 96 frames of 10 ms each.

       We repeat ground_truth and identifiers to fit the number of extracted VGGish features.
    """
    # 1) Extract log-mel spectrograms
    first_audio = True
    for p in paths:
        if first_audio:
            input_data = vggish_input.wavfile_to_examples(config['audio_folder'] + p)
            ground_truth = np.repeat(path2gt[p], input_data.shape[0], axis=0)
            identifiers = np.repeat(p, input_data.shape[0], axis=0)
            first_audio = False
        else:
            tmp_in = vggish_input.wavfile_to_examples(config['audio_folder'] + p)
            input_data = np.concatenate((input_data, tmp_in), axis=0)
            tmp_gt = np.repeat(path2gt[p], tmp_in.shape[0], axis=0)
            ground_truth = np.concatenate((ground_truth, tmp_gt), axis=0)
            tmp_id = np.repeat(p, tmp_in.shape[0], axis=0)
            identifiers = np.concatenate((identifiers, tmp_id), axis=0)

    # 2) Load Tensorflow model to extract VGGish features
    with tf.Graph().as_default(), tf.Session() as sess:
        vggish_slim.define_vggish_slim(training=False)
        vggish_slim.load_vggish_slim_checkpoint(sess, 'vggish_model.ckpt')
        features_tensor = sess.graph.get_tensor_by_name(vggish_params.INPUT_TENSOR_NAME)
        embedding_tensor = sess.graph.get_tensor_by_name(vggish_params.OUTPUT_TENSOR_NAME)
        extracted_feat = sess.run([embedding_tensor], feed_dict={features_tensor: input_data})
        feature = np.squeeze(np.asarray(extracted_feat))

    return [feature, ground_truth, identifiers] 

def extract_audioset_embedding():
    """Extract log mel spectrogram features. 
    """
    
    # Arguments & parameters
    mel_bins = vggish_params.NUM_BANDS
    sample_rate = vggish_params.SAMPLE_RATE
    input_len = vggish_params.NUM_FRAMES
    embedding_size = vggish_params.EMBEDDING_SIZE
    
    '''You may modify the EXAMPLE_HOP_SECONDS in vggish_params.py to change the 
    hop size. '''

    # Paths
    audio_path = 'appendixes/01.wav'
    checkpoint_path = os.path.join('vggish_model.ckpt')
    pcm_params_path = os.path.join('vggish_pca_params.npz')
    
    if not os.path.isfile(checkpoint_path):
        raise Exception('Please download vggish_model.ckpt from '
            'https://storage.googleapis.com/audioset/vggish_model.ckpt '
            'and put it in the root of this codebase. ')
        
    if not os.path.isfile(pcm_params_path):
        raise Exception('Please download pcm_params_path from '
        'https://storage.googleapis.com/audioset/vggish_pca_params.npz '
        'and put it in the root of this codebase. ')
    
    # Load model
    sess = tf.Session()
    
    vggish_slim.define_vggish_slim(training=False)
    vggish_slim.load_vggish_slim_checkpoint(sess, checkpoint_path)
    features_tensor = sess.graph.get_tensor_by_name(vggish_params.INPUT_TENSOR_NAME)
    embedding_tensor = sess.graph.get_tensor_by_name(vggish_params.OUTPUT_TENSOR_NAME)
    
    pproc = vggish_postprocess.Postprocessor(pcm_params_path)

    # Read audio
    (audio, _) = read_audio(audio_path, target_fs=sample_rate)
    
    # Extract log mel feature
    logmel = vggish_input.waveform_to_examples(audio, sample_rate)

    # Extract embedding feature
    [embedding_batch] = sess.run([embedding_tensor], feed_dict={features_tensor: logmel})
    
    # PCA
    postprocessed_batch = pproc.postprocess(embedding_batch)
    
    print('Audio length: {}'.format(len(audio)))
    print('Log mel shape: {}'.format(logmel.shape))
    print('Embedding feature shape: {}'.format(postprocessed_batch.shape))