Thanks for tuning in to Google I/O. View all sessions on demandWatch on demand

TensorFlow I/O

import tensorflow as tf
import tensorflow_io as tfio
path = tf.keras.utils.get_file('setero.wav', '')

# Open the file as a ``
audio_ds = tfio.IODataset.from_audio(path)
# => <AudioIODataset shapes: (2,), types: tf.int16>

# Or as a virtual Tensor that loads what's needed.
audio_io = tfio.IOTensor.from_audio(path)
# =>  <AudioIOTensor: spec=TensorSpec(shape=(880640, 2), dtype=tf.int16, name=None), rate=44100>
slice = audio_io[:1024]

TensorFlow I/O 是一系列的檔案系統和檔案格式,可用來補充 TensorFlow 的內建支援。

TensorFlow I/O 是由 TensorFlow SIG-IO 維護,提供實用的額外資料集、串流和檔案系統擴充功能。