Skip to content

Releases: TensorSpeech/TensorFlowASR

TensorFlowASR v0.5.0

17 Dec 16:00
644d5af
Compare
Choose a tag to compare

Release Notes

  • Supported ContextNet http://arxiv.org/abs/2005.03191
  • Updated train dataset
  • Updated example scripts
  • Add tf.sequence_mask into Transducer Prediction Network while training to mask padding values

TensorFlowASR v0.4.5

15 Dec 16:29
Compare
Choose a tag to compare

Release Notes

  • Add github publish workflow

TensorFlowASR v0.4.4

15 Dec 16:17
84017e3
Compare
Choose a tag to compare

Release Notes

  • Update transducer beam search to take norm_score
  • Update base tester

TensorFlowASR v0.4.3

12 Dec 17:09
fa97b4f
Compare
Choose a tag to compare

Release Notes

  • Fix gammatone feature extraction for using in tf.function
  • Add support for masking

TensorFlowASR v0.4.2

12 Dec 06:58
Compare
Choose a tag to compare

Release Notes

  • Update audio read to convert from multi channels into single channel

TensorFlowASR v0.4.1

09 Dec 17:40
Compare
Choose a tag to compare

Release Notes

  • Reformat code and fix minor bugs

TensorFlowASR v0.4.0

06 Dec 11:12
a9b0850
Compare
Choose a tag to compare

Release Notes

  • Updated RNN Transducer Beam Search from pure python operation to tf operations
  • Updated Test runner to accept batch size > 1
  • Updated Test datasets to return signals instead of speech features
  • Updated Conformer's Model Config names
  • Updated recognize functions to accept signals instead of speech features
  • Updated readme files.

TensorFlowASR v0.3.2

04 Dec 17:11
Compare
Choose a tag to compare

Release Notes

  • Fix minor bugs in trainer

TensorFlowASR v0.3.1

22 Nov 11:01
b42d6d4
Compare
Choose a tag to compare

Release Notes

  • Add additional_properties fields to config, in order for users to use their own extra configs
  • Remove tensorflow>=2.3.0 required dependency for more flexible installation

TensorFlowASR v0.3.0

14 Nov 16:12
1785335
Compare
Choose a tag to compare

Release Notes

  • Fix conformer tflite on tf-nightly
  • Supported Gradients Accumulation for Training in Larger Batch Size