Merge pull request tensorflow#2246 from zheng-xq/branch_121636618

Branch 121636618

Author: martinwicke

tensorflow/BUILD

@@ -73,6 +73,7 @@ filegroup(
         ":all_files",
         "//tensorflow/cc:all_files",
         "//tensorflow/contrib:all_files",
+        "//tensorflow/contrib/copy_graph:all_files",
         "//tensorflow/contrib/ctc:all_files",
         "//tensorflow/contrib/distributions:all_files",
         "//tensorflow/contrib/ffmpeg:all_files",

tensorflow/contrib/BUILD

@@ -14,6 +14,7 @@ py_library(
     visibility = ["//visibility:public"],
     deps = [
         "//tensorflow/contrib/bayesflow:bayesflow_py",
+        "//tensorflow/contrib/copy_graph:copy_graph_py",
         "//tensorflow/contrib/ctc:ctc_py",
         "//tensorflow/contrib/distributions:distributions_py",
         "//tensorflow/contrib/ffmpeg:ffmpeg_ops_py",
@@ -31,7 +32,6 @@ py_library(
         "//tensorflow/contrib/tensor_forest:tensor_forest_py",
         "//tensorflow/contrib/testing:testing_py",
         "//tensorflow/contrib/util:util_py",
-        "//tensorflow/contrib/copy_graph:copy_graph_py",
     ],
 )
 

tensorflow/contrib/copy_graph/python/__init__.py

@@ -12,4 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ==============================================================================
+"""Functions for copying elements from one graph to another.
+"""
 
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function

tensorflow/contrib/copy_graph/python/util/__init__.py

@@ -12,4 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ==============================================================================
+"""Functions for copying elements from one graph to another.
+"""
 
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function

tensorflow/contrib/ffmpeg/BUILD

@@ -31,10 +31,25 @@ cc_library(
     alwayslink = 1,
 )
 
+cc_library(
+    name = "encode_audio_op_cc",
+    srcs = ["encode_audio_op.cc"],
+    copts = tf_copts(),
+    linkstatic = 1,
+    visibility = ["//visibility:private"],
+    deps = [
+        "//third_party/eigen3",
+        "//tensorflow/contrib/ffmpeg/kernels:ffmpeg_lib",
+        "//tensorflow/core:framework_headers_lib",
+    ],
+    alwayslink = 1,
+)
+
 tf_custom_op_library(
-    name = "decode_audio_op.so",
+    name = "ffmpeg.so",
     deps = [
         ":decode_audio_op_cc",
+        ":encode_audio_op_cc",
     ],
 )
 
@@ -47,6 +62,15 @@ tf_gen_op_wrapper_py(
     ],
 )
 
+tf_gen_op_wrapper_py(
+    name = "encode_audio_op_py",
+    require_shape_functions = True,
+    visibility = ["//visibility:private"],
+    deps = [
+        ":encode_audio_op_cc",
+    ],
+)
+
 tf_py_test(
     name = "decode_audio_op_test",
     srcs = ["decode_audio_op_test.py"],
@@ -56,7 +80,25 @@ tf_py_test(
         "//tensorflow/python:platform",
     ],
     data = [
-        ":decode_audio_op.so",
+        ":ffmpeg.so",
+        ":test_data",
+    ],
+    tags = [
+        "local",
+        "manual",
+    ],
+)
+
+tf_py_test(
+    name = "encode_audio_op_test",
+    srcs = ["encode_audio_op_test.py"],
+    additional_deps = [
+        ":ffmpeg_ops_py",
+        "//third_party/py/tensorflow",
+        "//tensorflow/python:platform",
+    ],
+    data = [
+        ":ffmpeg.so",
         ":test_data",
     ],
     tags = [
@@ -72,11 +114,12 @@ py_library(
         "ffmpeg_ops.py",
     ],
     data = [
-        ":decode_audio_op.so",
+        ":ffmpeg.so",
     ],
     srcs_version = "PY2AND3",
     deps = [
         ":decode_audio_op_py",
+        ":encode_audio_op_py",
     ],
 )
 

tensorflow/contrib/ffmpeg/__init__.py

@@ -19,3 +19,4 @@
 from __future__ import print_function
 
 from tensorflow.contrib.ffmpeg.ffmpeg_ops import decode_audio
+from tensorflow.contrib.ffmpeg.ffmpeg_ops import encode_audio

tensorflow/contrib/ffmpeg/encode_audio_op.cc

@@ -0,0 +1,106 @@
+// Copyright 2016 Google Inc. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+// =============================================================================
+
+#include <limits>
+
+#include "tensorflow/contrib/ffmpeg/kernels/ffmpeg_lib.h"
+#include "tensorflow/core/framework/op.h"
+#include "tensorflow/core/framework/op_kernel.h"
+
+namespace tensorflow {
+namespace ffmpeg {
+
+class EncodeAudioOp : public OpKernel {
+ public:
+  explicit EncodeAudioOp(OpKernelConstruction* context)
+      : OpKernel(context) {
+    OP_REQUIRES_OK(context, context->GetAttr("file_format", &file_format_));
+    file_format_ = str_util::Lowercase(file_format_);
+    OP_REQUIRES(context, file_format_ == "wav",
+                errors::InvalidArgument("file_format arg must be \"wav\"."));
+
+    OP_REQUIRES_OK(
+        context, context->GetAttr("samples_per_second", &samples_per_second_));
+    OP_REQUIRES(context, samples_per_second_ > 0,
+                errors::InvalidArgument("samples_per_second must be > 0."));
+  }
+
+  void Compute(OpKernelContext* context) override {
+    // Get and verify the input data.
+    OP_REQUIRES(context, context->num_inputs() == 1,
+                errors::InvalidArgument(
+                    "EncodeAudio requires exactly one input."));
+    const Tensor& contents = context->input(0);
+    OP_REQUIRES(context, TensorShapeUtils::IsMatrix(contents.shape()),
+                errors::InvalidArgument(
+                    "sampled_audio must be a rank 2 tensor but got shape ",
+                    contents.shape().DebugString()));
+    OP_REQUIRES(
+        context, contents.NumElements() <= std::numeric_limits<int32>::max(),
+        errors::InvalidArgument(
+            "sampled_audio cannot have more than 2^31 entries. Shape = ",
+            contents.shape().DebugString()));
+
+    // Create the encoded audio file.
+    std::vector<float> samples;
+    samples.reserve(contents.NumElements());
+    for (int32 i = 0; i < contents.NumElements(); ++i) {
+      samples.push_back(contents.flat<float>()(i));
+    }
+    const int32 channel_count = contents.dim_size(1);
+    string encoded_audio;
+    OP_REQUIRES_OK(context,
+                   CreateAudioFile(file_format_, samples_per_second_,
+                                   channel_count, samples, &encoded_audio));
+
+    // Copy the encoded audio file to the output tensor.
+    Tensor* output = nullptr;
+    OP_REQUIRES_OK(context,
+                   context->allocate_output(0, TensorShape(), &output));
+    output->scalar<string>()() = encoded_audio;
+  }
+
+ private:
+  string file_format_;
+  int32 samples_per_second_;
+};
+
+REGISTER_KERNEL_BUILDER(Name("EncodeAudio").Device(DEVICE_CPU), EncodeAudioOp);
+
+REGISTER_OP("EncodeAudio")
+    .Input("sampled_audio: float")
+    .Output("contents: string")
+    .Attr("file_format: string")
+    .Attr("samples_per_second: int")
+    .Doc(R"doc(
+Processes a `Tensor` containing sampled audio with the number of channels
+and length of the audio specified by the dimensions of the `Tensor`. The
+audio is converted into a string that, when saved to disk, will be equivalent
+to the audio in the specified audio format.
+
+The input audio has one row of the tensor for each channel in the audio file.
+Each channel contains audio samples starting at the beginning of the audio and
+having `1/samples_per_second` time between them. The output file will contain
+all of the audio channels contained in the tensor.
+
+sampled_audio: A rank 2 tensor containing all tracks of the audio. Dimension 0
+    is time and dimension 1 is the channel.
+contents: The binary audio file contents.
+file_format: A string describing the audio file format. This must be "wav".
+samples_per_second: The number of samples per second that the audio should have.
+)doc");
+
+}  // namespace ffmpeg
+}  // namespace tensorflow

tensorflow/contrib/ffmpeg/encode_audio_op_test.py

@@ -0,0 +1,50 @@
+# Copyright 2016 Google Inc. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# =============================================================================
+
+"""Tests for third_party.tensorflow.contrib.ffmpeg.encode_audio_op."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os.path
+
+import tensorflow as tf
+
+from tensorflow.contrib import ffmpeg
+from tensorflow.python.platform import resource_loader
+
+
+class EncodeAudioOpTest(tf.test.TestCase):
+
+  def testRoundTrip(self):
+    """Fabricates some audio, creates a wav file, reverses it, and compares."""
+    with self.test_session():
+      path = os.path.join(
+          resource_loader.get_data_files_path(), 'testdata/mono_10khz.wav')
+      with open(path, 'r') as f:
+        original_contents = f.read()
+
+      audio_op = ffmpeg.decode_audio(
+          original_contents, file_format='wav', samples_per_second=10000,
+          channel_count=1)
+      encode_op = ffmpeg.encode_audio(
+          audio_op, file_format='wav', samples_per_second=10000)
+      encoded_contents = encode_op.eval()
+      self.assertEqual(original_contents, encoded_contents)
+
+
+if __name__ == '__main__':
+  tf.test.main()

tensorflow/contrib/ffmpeg/ffmpeg_ops.py

@@ -20,6 +20,7 @@
 from __future__ import print_function
 
 from tensorflow.contrib.ffmpeg.ops import gen_decode_audio_op_py
+from tensorflow.contrib.ffmpeg.ops import gen_encode_audio_op_py
 from tensorflow.python.framework import load_library
 from tensorflow.python.framework import ops
 from tensorflow.python.framework import tensor_shape
@@ -75,6 +76,39 @@ def decode_audio(contents, file_format=None, samples_per_second=None,
 ops.NoGradient('DecodeAudio')
 
 
+@ops.RegisterShape('EncodeAudio')
+def _encode_audio_shape(unused_op):
+  """Computes the shape of an EncodeAudio operation.
+
+  Returns:
+    A list of output shapes. There's exactly one output, the formatted audio
+    file. This is a rank 0 tensor.
+  """
+  return [tensor_shape.TensorShape([])]
+
+
+def encode_audio(audio, file_format=None, samples_per_second=None):
+  """Creates an op that encodes an audio file using sampled audio from a tensor.
+
+  Args:
+    audio: A rank 2 tensor that has time along dimension 0 and channels along
+        dimension 1. Dimension 0 is `samples_per_second * length` long in
+        seconds.
+    file_format: The type of file to encode. "wav" is the only supported format.
+    samples_per_second: The number of samples in the audio tensor per second of
+        audio.
+
+  Returns:
+    A scalar tensor that contains the encoded audio in the specified file
+    format.
+  """
+  return gen_encode_audio_op_py.encode_audio(
+      audio, file_format=file_format, samples_per_second=samples_per_second)
+
+
+ops.NoGradient('EncodeAudio')
+
+
 def _load_library(name, op_list=None):
   """Loads a .so file containing the specified operators.
 
@@ -97,4 +131,4 @@ def _load_library(name, op_list=None):
                       (expected_op, name))
 
 
-_load_library('decode_audio_op.so', ['DecodeAudio'])
+_load_library('ffmpeg.so', ['DecodeAudio', 'EncodeAudio'])

tensorflow/contrib/ffmpeg/kernels/ffmpeg_lib.h

@@ -29,8 +29,10 @@ string GetTempFilename(const string& extension);
 // Reads an audio file using ffmpeg and converts it into an array of samples in
 // [-1.0, 1.0]. If there are multiple channels in the audio then each frame will
 // contain a separate sample for each channel. Frames are ordered by time.
-Status ReadAudioFile(const string& filename, const string& audio_format_id,
-                     int32 samples_per_second, int32 channel_count,
+Status ReadAudioFile(const string& filename,
+                     const string& audio_format_id,
+                     int32 samples_per_second,
+                     int32 channel_count,
                      std::vector<float>* output_samples);
 
 // Creates an audio file using ffmpeg in a specific format. The samples are in