Skip to content
This repository was archived by the owner on Jul 7, 2023. It is now read-only.

1.6.2 #771

Merged
merged 12 commits into from
May 8, 2018
Merged

1.6.2 #771

Changes from 1 commit
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
b7e5f48
Adding Lambada dataset as a new problem to T2T.
a-googler May 4, 2018
0d8f6f6
Enable bfloat16 activation for Image Transformer model.
a-googler May 4, 2018
ed0dc39
Zero-pad speech utterances to bring closer training (with zero paddin…
a-googler May 6, 2018
5a0e7af
Creation of ASR with transformer demo notebook.
a-googler May 7, 2018
a40ffb3
Using _slow_greedy_infer and _beam_decode_slow in _greey_infer and _b…
a-googler May 7, 2018
edf4e05
Removed the following from transformer_vae :
May 7, 2018
3c53555
Internal decoding merge and update Travis
May 7, 2018
2ee03ca
Allow for decompress outside do_mask. It seems to be independent of t…
May 7, 2018
b748dbe
Corrections to autoencoders and atari ROM paths.
May 8, 2018
59e4a82
Bump version and link ASR notebook and tutorial.
May 8, 2018
6be28ff
Update base hparams for tpu
May 8, 2018
485e5d9
internal
May 8, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Zero-pad speech utterances to bring closer training (with zero paddin… 
…g from batching) and testing conditions

PiperOrigin-RevId: 195565730
  • Loading branch information
T2T Team authored and lukaszkaiser committed May 8, 2018
commit ed0dc39af97c174525e2a7ece09197aec2252b32
6 changes: 4 additions & 2 deletions tensor2tensor/data_generators/speech_recognition.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -251,6 +251,7 @@ def hparams(self, defaults, model_hparams):
p.add_hparam("audio_upper_edge_hertz", 8000.0)
p.add_hparam("audio_num_mel_bins", 80)
p.add_hparam("audio_add_delta_deltas", True)
p.add_hparam("num_zeropad_frames", 250)

p = defaults
# p.stop_at_eos = int(False)
Expand Down Expand Up @@ -319,8 +320,9 @@ def preprocess_example(self, example, mode, hparams):

# Later models like to flatten the two spatial dims. Instead, we add a
# unit spatial dim and flatten the frequencies and channels.
example["inputs"] = tf.reshape(
mel_fbanks, [fbank_size[1], fbank_size[2], fbank_size[3]])
example["inputs"] = tf.concat([
tf.reshape(mel_fbanks, [fbank_size[1], fbank_size[2], fbank_size[3]]),
tf.zeros((p.num_zeropad_frames, fbank_size[2], fbank_size[3]))], 0)

if not p.audio_keep_example_waveforms:
del example["waveforms"]
Expand Down