Merge pull request #98 from didi/compat

Compat tf v1.x version

Author: zh794390558

.travis.yml

@@ -6,8 +6,9 @@ services:
 before_install:
   - export DELTA_PATH=`pwd`; echo $DELTA_PATH
   - export DOCKER_DELTA="/home/gitlab-runner/delta"; echo $DOCKER_DELTA
-  - docker pull zh794390558/delta:1.14.0-ci-cpu-py3
-  - docker run -it -d --name travis_con --user root -v ${DELTA_PATH}:${DOCKER_DELTA} zh794390558/delta:1.14.0-ci-cpu-py3 bash
+  - export CI_IMAGE=zh794390558/delta:2.0.0-ci-cpu-py3
+  - docker pull ${CI_IMAGE}
+  - docker run -it -d --name travis_con --user root -v ${DELTA_PATH}:${DOCKER_DELTA} ${CI_IMAGE} bash
   - docker exec travis_con bash -c "gcc -v && g++ -v"
   - docker exec travis_con bash -c "cd ${DOCKER_DELTA}; source env.sh"
   - docker exec travis_con bash -c "cd ${DOCKER_DELTA}/tools; touch test.done"

README.md

@@ -78,9 +78,9 @@ To verify the installation, run:
 
 ```shell
 # Activate conda environment
-conda activate delta-py3.6-tf1.14
+conda activate delta-py3.6-tf2.0.0
 # Or use the following command if your conda version is < 4.6
-# source activate delta-py3.6-tf1.14
+# source activate delta-py3.6-tf2.0.0
 
 # Add DELTA enviornment
 source env.sh

delta/data/feat/speech_feature.py

@@ -17,7 +17,7 @@
 import os
 
 import numpy as np
-import tensorflow as tf
+import delta.compat as tf
 from absl import logging
 
 from delta.data.feat import speech_ops

delta/data/feat/speech_feature_test.py

@@ -19,7 +19,7 @@
 
 import librosa
 import numpy as np
-import tensorflow as tf
+import delta.compat as tf
 from absl import logging
 
 from delta.data.feat import speech_ops

delta/data/feat/speech_ops.py

@@ -18,10 +18,11 @@
 from absl import logging
 
 #pylint: disable=no-name-in-module
-import tensorflow as tf
-from tensorflow.contrib.framework.python.ops import audio_ops as contrib_audio
+import delta.compat as tf
+from tensorflow.python.ops import gen_audio_ops as audio_ops
 
 from delta import utils
+from delta.utils.hparam import HParams
 from delta.layers.ops import py_x_ops
 
 
@@ -36,7 +37,7 @@ def speech_params(sr=16000,
                   cmvn=False,
                   cmvn_path=''):
   ''' speech feat params '''
-  p = tf.contrib.training.HParams()
+  p = HParams()
   p.add_hparam("audio_sample_rate", sr)
   if dither:
     p.add_hparam("audio_dither", 1.0 / np.iinfo(np.int16).max)
@@ -68,7 +69,7 @@ def read_wav(wavfile, params):
   '''
   contents = tf.read_file(wavfile)
   #pylint: disable=no-member
-  waveforms = contrib_audio.decode_wav(
+  waveforms = tf.audio.decode_wav(
       contents,
       desired_channels=params.audio_desired_channels,
       desired_samples=params.audio_desired_samples,
@@ -103,7 +104,7 @@ def powspec_feat(samples,
   del preemph
 
   #pylint: disable=no-member
-  feat = contrib_audio.audio_spectrogram(
+  feat = audio_ops.audio_spectrogram(
       samples,
       window_size=winlen * sr,
       stride=winstep * sr,
@@ -303,7 +304,7 @@ def _loop_body(time, inputs, output_tas):
   loop_vars = (time, waveforms, output_tas)
 
   parallel_iterations = 10
-  shape_invariants = tf.contrib.framework.nest.map_structure(
+  shape_invariants = tf.nest.map_structure(
       lambda t: tf.TensorShape(None), loop_vars)
 
   (time, inputs, output_tas) = tf.while_loop(
@@ -361,7 +362,7 @@ def _loop_body(time, end_time, context, left_context, right_context,
     loop_vars = (time, T, context, left_context, right_context, output_tas)
 
     parallel_iterations = 10
-    shape_invariants = tf.contrib.framework.nest.map_structure(
+    shape_invariants = tf.nest.map_structure(
         lambda t: tf.TensorShape(None), loop_vars)
 
     (time, end_time, context, left_context, right_context,

delta/data/feat/speech_ops_test.py

@@ -17,7 +17,7 @@
 import os
 from pathlib import Path
 import numpy as np
-import tensorflow as tf
+import delta.compat as tf
 from absl import logging
 
 from delta.data.feat.speech_feature import load_wav

delta/data/feat/tf_speech_feature.py

@@ -17,12 +17,12 @@
 import functools
 
 import numpy as np
-import tensorflow as tf
+import delta.compat as tf
 import scipy.signal
 #pylint: disable=ungrouped-imports,no-name-in-module
-from tensorflow.contrib.framework.python.ops import audio_ops as contrib_audio
 
 from delta import utils
+from delta.utils.hparam import HParams
 
 
 def add_delta_deltas(filterbanks, name=None):
@@ -61,7 +61,7 @@ def compute_mel_filterbank_features(waveforms,
                                     frame_step=10,
                                     fft_length=None,
                                     window_fn=functools.partial(
-                                        tf.contrib.signal.hann_window,
+                                        tf.signal.hann_window,
                                         periodic=True),
                                     lower_edge_hertz=80.0,
                                     upper_edge_hertz=7600.0,
@@ -109,7 +109,7 @@ def compute_mel_filterbank_features(waveforms,
   if fft_length is None:
     fft_length = int(2**(np.ceil(np.log2(frame_length))))
 
-  stfts = tf.contrib.signal.stft(
+  stfts = tf.signal.stft(
       waveforms,
       frame_length=frame_length,
       frame_step=frame_step,
@@ -130,7 +130,7 @@ def compute_mel_filterbank_features(waveforms,
   # Warp the linear-scale, magnitude spectrograms into the mel-scale.
   num_spectrogram_bins = magnitude_spectrograms.shape[-1].value
   linear_to_mel_weight_matrix = (
-      tf.contrib.signal.linear_to_mel_weight_matrix(num_mel_bins,
+      tf.signal.linear_to_mel_weight_matrix(num_mel_bins,
                                                     num_spectrogram_bins,
                                                     sample_rate,
                                                     lower_edge_hertz,
@@ -153,7 +153,7 @@ def read_wav(wavfile, params):
   ''' samples of shape [nsample] '''
   contents = tf.read_file(wavfile)
   #pylint: disable=no-member
-  waveforms = contrib_audio.decode_wav(
+  waveforms = tf.audio.decode_wav(
       contents,
       desired_channels=params.audio_channels,
       #desired_samples=params.audio_sample_rate,
@@ -176,7 +176,7 @@ def speech_params(sr=16000,
                   cmvn=False,
                   cmvn_path=''):
   ''' feat params '''
-  p = tf.contrib.training.HParams()
+  p = HParams()
   p.add_hparam("audio_sample_rate", sr)
   p.add_hparam("audio_channels", 1)
   p.add_hparam("audio_preemphasis", 0.97)

delta/data/feat/tf_speech_feature_test.py

@@ -17,7 +17,7 @@
 import os
 from pathlib import Path
 
-import tensorflow as tf
+import delta.compat as tf
 from absl import logging
 
 from delta.data.feat.speech_feature import load_wav

delta/data/frontend/analyfiltbank.py

@@ -14,7 +14,7 @@
 # limitations under the License.
 # ==============================================================================
 
-import tensorflow as tf
+import delta.compat as tf
 
 from delta.layers.ops import py_x_ops
 from delta.utils.hparam import HParams
@@ -67,7 +67,7 @@ def call(self, audio_data, sample_rate):
       if sample_rate == None:
         sample_rate = tf.constant(p.sample_rate, dtype=float)
 
-      assert_op = tf.compat.v1.assert_equal(
+      assert_op = tf.assert_equal(
           tf.constant(p.sample_rate), tf.cast(sample_rate, dtype=float))
       with tf.control_dependencies([assert_op]):
 

delta/data/frontend/analyfiltbank_test.py

@@ -14,7 +14,7 @@
 # limitations under the License.
 # ==============================================================================
 
-import tensorflow as tf
+import delta.compat as tf
 import os
 from pathlib import Path
 from delta.data.frontend.read_wav import ReadWav