over_lap_and_add

colab版本
from keras.layers.normalization.batch_normalization_v1 import BatchNormalization
本地版
from keras.layers.normalization.batch_normalization_v1 import BatchNormalization

    def over_lap_and_add(self, framed_signals, frame_length, frame_step, winfunc=tf.signal.hamming_window):
      """overlap and add
      params：
        framed_signals: tf.float32, shape=[batch, n_frames, frame_length]
        frame_length: Window length
        frame_step: frame shift
      return：
        signals: tf.float32, shape=[batch, x_length]
      """
      shape = tf.shape(framed_signals)
      n_frames = shape[1]
      # Generate de-overlapping windows
      if winfunc is not None:
        window = winfunc(frame_length, dtype=tf.float32)
        window = tf.reshape(window, [1, frame_length])
        window = tf.tile(window, [n_frames, 1])
        window = tf.signal.overlap_and_add(window, frame_step)
      signals = tf.signal.overlap_and_add(framed_signals, frame_step)
      signals /= window
      signals = tf.cast(signals, tf.float32)
      return signals

SNR