v0.5.0-alpha.5

Merge pull request #2019 from lissyx/bump-v0.5.0-alpha-5

Bump VERSION to 0.5.0-alpha.5

v0.5.0-alpha.4

Merge pull request #1969 from lissyx/bump-v0.5.0-alpha.4

Bump VERSION to v0.5.0-alpha.4

v0.5.0-alpha.3

Merge pull request #1966 from lissyx/bump-v0.5.0-alpha.3

Bump VERSION to v0.5.0-alpha.3

v0.5.0-alpha.2

Merge pull request #1951 from lissyx/force-rebuild

Force rebuild

v0.5.0-alpha.1

Merge pull request #1851 from lissyx/bump-v0.5.0-alpha.1

Bump VERSION to 0.5.0-alpha.1

v0.5.0-alpha.0

Merge pull request #1846 from lissyx/bump-v0.5.0-alpha.0

Bump VERSION to 0.5.0-alpha.0

Deep Speech 0.4.1

General

This is the 0.4.1 release of Deep Speech, an open speech-to-text engine. This release includes source code

v0.4.1.tar.gz

and a trained model

deepspeech-0.4.1-models.tar.gz

trained on American English which achieves an 8.26% word error rate on the LibriSpeech clean test corpus (models with "rounded" in their file name have rounded weights and those with a "*.pbmm" extension are memory mapped and much more memory efficient), and example audio

audio-0.4.1.tar.gz

which can be used to test the engine and checkpoint files

deepspeech-0.4.1-checkpoint.tar.gz

which can be used as the basis for further fine-tuning.

Notable changes from the previous release

• Removed initialize_from_frozen_model flag and support code (46d1cec)
• Updated to TensorFlow r1.12 (26f7bc3, d243839)
• Switched to CTC algorithm from TensorFlow's to to ctcdecode (#1679, #1693, #1696)
• Remove old AOT model (1540fa3)
• Introduced NodeJS v11.x support (5536e54, 4867d6b)
• Directly exporting TFLite ready model (5d30afd, d6642da)
• Added streaming API Support to the GUI Tool (851fb4e)
• Fixed Fisher + Switchboard importers (9aa23ed, f036691, b37f1a7, b2f967a, 6c903c8, ba62289, 81b1600)
• Fixes in Switchboard importer (8507f99, e47e344, 4efc5c6)
• Add example for Python streaming from mic with VAD (74cebb8)
• Fixed MFCC window size + stride (1df9602, 2a8128b)
• Renamed LM_WEIGHT and VALID_WORD_COUNT_WEIGHT to LM_ALPHA and LM_BETA respectively and changed their values (fc46f43)

Hyperparameters for fine-tuning

The hyperparameters used to train the model are useful for fine tuning. Thus, we document them here along with the hardware used, a server with 8 TitanX Pascal GPUs (12GB of VRAM).

• train_files Fisher, LibriSpeech, Switchboard training corpora, as well as a pre-release snapshot of the English Common Voice training corpus.
• dev_files LibriSpeech clean and other dev corpora, as well as a pre-release snapshot of the English Common Voice validation corpus.
• test_files LibriSpeech clean test corpus
• train_batch_size 24
• dev_batch_size 48
• test_batch_size 48
• epoch 30
• learning_rate 0.0001
• display_step 0
• validation_step 1
• dropout_rate 0.15
• checkpoint_step 1
• n_hidden 2048
• lm_alpha 0.75
• lm_beta 1.85

The weights with the best validation loss were selected at the end of the 30 epochs.

Bindings

This release also includes a Python based command line tool deepspeech, installed through

pip install deepspeech

Alternatively, quicker inference can be performed using a supported NVIDIA GPU on Linux. (See below to find which GPU's are supported.) This is done by instead installing the GPU specific package:

pip install deepspeech-gpu

Also, it exposes bindings for the following languages

• Python (Versions 2.7, 3.4, 3.5, 3.6 and 3.7) installed via

  pip install deepspeech

  Alternatively, quicker inference can be performed using a supported NVIDIA GPU on Linux. (See below to find which GPU's are supported.) This is done by instead installing the GPU specific package:

  pip install deepspeech-gpu

• NodeJS (Versions 4.x, 5.x, 6.x, 7.x, 8.x, 9.x, 10.x, and 11.x) installed via

  npm install deepspeech
  

  Alternatively, quicker inference can be performed using a supported NVIDIA GPU on Linux. (See below to find which GPU's are supported.) This is done by instead installing the GPU specific package:

  npm install deepspeech-gpu
  

• C++ which requires the appropriate shared objects are installed from native_client.tar.xz (See the section in the main README which describes native_client.tar.xz installation.)

In addition there are third party bindings that are supported by external developers, for example

• Rust which is installed by following the instructions on the external Rust repo.
• Go which is installed by following the instructions on the external Go repo.

Supported Platforms

• OS X 10.10, 10.11, 10.12, 10.13 and 10.14
• Linux x86 64 bit with a modern CPU (Needs at least AVX/FMA)
• Linux x86 64 bit with a modern CPU + NVIDIA GPU (Compute Capability at least 3.0, see NVIDIA docs)
• Raspbian Stretch on Raspberry Pi 3
• ARM64 built against Debian/ARMbian Stretch and tested on LePotato boards
• Java Android bindings / demo app. Early preview, tested only on Pixel 2 device, TF Lite model only

Known Issues

• Feature caching speeds training but increases memory usage
• Current v2 TRIE handling still triggers ~600MB memory usage

Contact/Getting Help

1. FAQ - We have a list of common questions, and their answers, in our FAQ. When just getting started, it's best to first check the FAQ to see if your question is addressed.
2. Discourse Forums - If your question is not addressed in the FAQ, the Discourse Forums is the next place to look. They contain conversations on General Topics, Using Deep Speech, Alternative Platforms, and Deep Speech Development.
3. IRC - If your question is not addressed by either the FAQ or Discourse Forums, you can contact us on the #machinelearning channel on Mozilla IRC; people there can try to answer/help
4. Issues - Finally, if all else fails, you can open an issue in our repo if there is a bug with the current code base.

Contributors to 0.4.1 release

• Reuben Morais
• Alexandre Lissy
• Kelly Davis
• Carlos Fonseca Murillo
• Rob Smith
• Igor Fritzsch
• Josh Meyer
• Nicolas Panel
• Benjamin Abel
• Vamsi Krishna Surapureddi
• b-ak
• David Zurow
• Célian Garcia
• Jegathesan Shanmugam
• Jahir Islam
• Jan Engelmohr
• Mikhail Kuznetcov

Deep Speech 0.4.0

General

This is the 0.4.0 release of Deep Speech, an open speech-to-text engine. This release includes source code

v0.4.0.tar.gz

and a trained model

deepspeech-0.4.0-models.tar.gz

trained on American English which achieves an 8.26% word error rate on the LibriSpeech clean test corpus (The incorrect model was uploaded this will be fixed in 0.4.1) (models with "rounded" in their file name have rounded weights and those with a "*.pbmm" extension are memory mapped and much more memory efficient), and example audio

audio-0.4.0.tar.gz

which can be used to test the engine and checkpoint files

deepspeech-0.4.0-checkpoint.tar.gz

which can be used as the basis for further fine-tuning.

Notable changes from the previous release

• Removed initialize_from_frozen_model flag and support code (46d1cec)
• Updated to TensorFlow r1.12 (26f7bc3, d243839)
• Switched to CTC algorithm from TensorFlow's to to ctcdecode (#1679, #1693, #1696)
• Remove old AOT model (1540fa3)
• Introduced NodeJS v11.x support (5536e54, 4867d6b)
• Directly exporting TFLite ready model (5d30afd, d6642da)
• Added streaming API Support to the GUI Tool (851fb4e)
• Fixed Fisher + Switchboard importers (9aa23ed, f036691, b37f1a7, b2f967a, 6c903c8, ba62289, 81b1600)
• Fixes in Switchboard importer (8507f99, e47e344, 4efc5c6)
• Add example for Python streaming from mic with VAD (74cebb8)
• Fixed MFCC window size + stride (1df9602, 2a8128b)

Hyperparameters for fine-tuning

The hyperparameters used to train the model are useful for fine tuning. Thus, we document them here along with the hardware used, a server with 8 TitanX Pascal GPUs (12GB of VRAM).

• train_files Fisher, LibriSpeech, Switchboard training corpora, as well as a pre-release snapshot of the English Common Voice training corpus.
• dev_files LibriSpeech clean and other dev corpora, as well as a pre-release snapshot of the English Common Voice validation corpus.
• test_files LibriSpeech clean test corpus
• train_batch_size 24
• dev_batch_size 48
• test_batch_size 48
• epoch 30
• learning_rate 0.0001
• display_step 0
• validation_step 1
• dropout_rate 0.15
• checkpoint_step 1
• n_hidden 2048
• lm_alpha 0.75
• lm_beta 1.85

The weights with the best validation loss were selected at the end of the 30 epochs.

Bindings

This release also includes a Python based command line tool deepspeech, installed through

pip install deepspeech

Alternatively, quicker inference can be performed using a supported NVIDIA GPU on Linux. (See below to find which GPU's are supported.) This is done by instead installing the GPU specific package:

pip install deepspeech-gpu

Also, it exposes bindings for the following languages

• Python (Versions 2.7, 3.4, 3.5, 3.6 and 3.7) installed via

  pip install deepspeech

  Alternatively, quicker inference can be performed using a supported NVIDIA GPU on Linux. (See below to find which GPU's are supported.) This is done by instead installing the GPU specific package:

  pip install deepspeech-gpu

• NodeJS (Versions 4.x, 5.x, 6.x, 7.x, 8.x, 9.x, 10.x, and 11.x) installed via

  npm install deepspeech
  

  Alternatively, quicker inference can be performed using a supported NVIDIA GPU on Linux. (See below to find which GPU's are supported.) This is done by instead installing the GPU specific package:

  npm install deepspeech-gpu
  

• C++ which requires the appropriate shared objects are installed from native_client.tar.xz (See the section in the main README which describes native_client.tar.xz installation.)

In addition there are third party bindings that are supported by external developers, for example

• Rust which is installed by following the instructions on the external Rust repo.
• Go

Supported Platforms

• OS X 10.10, 10.11, 10.12, 10.13 and 10.14
• Linux x86 64 bit with a modern CPU (Needs at least AVX/FMA)
• Linux x86 64 bit with a modern CPU + NVIDIA GPU (Compute Capability at least 3.0, see NVIDIA docs)
• Raspbian Stretch on Raspberry Pi 3
• ARM64 built against Debian/ARMbian Stretch and tested on LePotato boards

Known Issues

• Feature caching speeds training but increases memory usage
• Current v2 TRIE handling still triggers ~600MB memory usage
• Incorrect model was uploaded to release which will be fixed in 0.4.1

Contact/Getting Help

1. FAQ - We have a list of common questions, and their answers, in our FAQ. When just getting started, it's best to first check the FAQ to see if your question is addressed.
2. Discourse Forums - If your question is not addressed in the FAQ, the Discourse Forums is the next place to look. They contain conversations on General Topics, Using Deep Speech, Alternative Platforms, and Deep Speech Development.
3. IRC - If your question is not addressed by either the FAQ or Discourse Forums, you can contact us on the #machinelearning channel on Mozilla IRC; people there can try to answer/help
4. Issues - Finally, if all else fails, you can open an issue in our repo if there is a bug with the current code base.

Contributors to 0.4.0 release

• Reuben Morais
• Alexandre Lissy
• Kelly Davis
• Carlos Fonseca Murillo
• Rob Smith
• Igor Fritzsch
• Josh Meyer
• Nicolas Panel
• Benjamin Abel
• Vamsi Krishna Surapureddi
• b-ak
• David Zurow
• Célian Garcia
• Jegathesan Shanmugam
• Jahir Islam
• Jan Engelmohr
• Mikhail Kuznetcov

v0.4.0-alpha.3

Merge pull request #1800 from lissyx/bump-v0.4.0-alpha.3

Bump VERSION to 0.4.0-alpha.3

v0.4.0-alpha.2

Merge pull request #1789 from lissyx/fix-nc-asset-name

Move nc_asset_name to extra