Dragonfire Versions Save

This is a security patch to fix Heap buffer overflow in UnsortedSegmentSum (CVE-2019-16778) vulnerability of TensorFlow package.

Changelog:

• Upgraded pip version from 9.0.1 to 19.3.1 for TensorFlow 1.15.0
• Upgraded TensorFlow version from 1.14.0 to 1.15.0
• Upgraded DeepPavlov version from 0.6.1 to 0.7.1

By integrating DeepPavlov SQuAD BERT into Dragonfire we have now more accurate Open-Domain Question Answering system without a need to have huge Wiki dumps or graph databases. Check out the ODQA Performance in this CI job.

Changelog:

• The badge generation service changed from Badgen to Shields
• Omniscient module replaced with DeepPavlov SQuAD BERT and renamed as ODQA
• Loglevel of TensorFlow set to ERROR and disabled FutureWarning(s)
• Added a GitHub workflow for automated Docker builds
• Travis CI is removed
• requests package version upgraded to >=2.20.0
• Post-installation scripts are merged
• The Docker image has been updated
• Added separate GitHub workflows to handle Automated Tests and Linter Checks
• Added a GitHub workflow to test the ODQA module's performance against HotpotQA dataset
• Fixed most of the major code maintainability issues detected by Code Climate
• Added a GitHub workflow to automatically publish a GitHub release, build and upload a Debian package to release assets
• Checksum values are updated in the post-installation script

In this release we have finally upgraded the TensorFlow version to 1.14.0 with the purpose of addressing the issue reports that tell us, it's impossible to install Dragonfire on relatively newer systems. We have also polished our CI pipelines to have more strict delivery cycles.

Changelog:

• Upgraded TensorFlow version from 1.0.0 to 1.14.0
• Upgraded spaCy version from 2.0.13 to 2.1.3
• Upgraded NeuralCoref version to 4.0
• Updated the DeepConversation Tensorflow model and published (v3)
• Refactored some of the NLP methods
• Updated the Docker image
• Added a Makefile to simplify installation commands
• Fixed some of the tests
• Fixed the CI pipelines
• Implemented code coverage by integrating CI to Codecov
• Integrated Code Climate to have automated code reviews and metrics
• Fixed Read the Docs documentation build pipeline

With the DeepSpeech's recently published new model in their Deep Speech 0.4.1 release the non-commercial speech recognition ability of Dragonfire is significantly increased.

Changelog:

• libgtk2.0-0 and gir1.2-gtk-3.0 dependencies are added
• The issue related to inactive lock in the speech recognizers is fixed
• deepspeech Python package dependency is upgraded from 0.2.0a5 to 0.4.1
• The DeepSpeech the model from 0.1.1 to 0.4.1
• Time telling feature(a built-in command) is added
• Several issues in built-in commands are fixed
• Release version mismatch is fixed

With this release Coreference resolution ability and Python 3.6 support comes to Dragonfire. Many thanks to NeuralCoref project for coreference resolution feature.

Changelog:

• Dependency issues are fixed
• Several minor bugs are fixed
• Python 3.6 support added
• Several linguistic mistakes are fixed
• A Dockerfile added
• Command-line formatting improved
• Add SQLite and MySQL support for the learning ability
• The knowledge base used in learning, migrated from TinyDB to SQLite/MySQL
• Module imports are fixed
• Automatic documentation generation added using Sphinx
• 134 different tests added and being checked with pytest using Travis CI
• Coreference resolution added using NeuralCoref model

After approximately 3 years of development, now we are finally ready to release version 1.0 of Dragonfire open-source virtual assistant.

Changelog:

• Post-installation scripts are fixed
• An API implemented
• The project is turned into more of a server-side application although it's continuing to preserve the features of its desktop application aspect
• The package is now Python 3 only
• Universal Python wheel generation disabled
• Basic Analyzer (done in dragonfire/__init__.py) is heavily pruned and improved. Now it's purely spaCy based
• Omniscient and Learner classes are refactored
• Learning ability now can store data (that comes from many different Android users) on MySQL database seamlessly
• Gspeech alternative added

The long waited Mozilla's DeepSpeech project finally released a pre-trained model on Jan 31, 2018 and we are proud to announce that Dragonfire is the first virtual assistant project that is successfully integrated Mozilla DeepSpeech into its speech recognition system. By gaining the advantage of DeepSpeech, Dragonfire is now able to understand your words with a much higher success rate (low WER results).

Changelog:

• Migrated from Kaldi Speech Recognition Toolkit to Mozilla DeepSpeech
• Python3 compatibility improved
• System tray icon is now GTK+ based instead of wxPython

With this release, Dragonfire is stable, reliable and more human-like than ever. Q&A and Learning capabilities are dramatically increased. So you will experience a more natural conversation with her.

Changelog:

• Python3 support added
• Omniscient Q&A Engine success rate is improved by ~40%
• The learning capabilities are improved by including the inverted case
• Wikipedia connection and disambiguation errors are fixed
• Several minor bugs are fixed
• Missing package dependencies are added
• Post-installation scripts are fixed

With the migration to Kaldi Speech Recognition Toolkit, Dragonfire is now truly independent from any kind of web service. That means she will be able recognize your speech without the need to an internet access.

Changelog:

• Migrated from PyPI SpeechRecognition package to Kaldi Speech Recognition Toolkit
• Migrated from festival cmu_us_clb_arctic voice to flite slt voice
• Kaldi GStreamer English model is added
• Google Search built-in commands are added
• Several minor bugs are fixed
• A number of experimental(non-functional) scripts are added