v1.0.6

Fix model download bug due to Flask update

v1.0.5

Changes:

• Updated dependencies, including Python version, Transformers, PyTorch, most packages.
• Minor changes to HFT model class in response to Transformers Trainer update.

v1.0.4

Added feature that allows:

• locally downloading models trained using Elpis.
• downloading fine-tuned or pre-trained models from HuggingFace which can then be further fine-tuned on particular user datasets.
• uploading trained models for the purposes of performing transcription.

Changes:

• Changed training dashboard layout.
• Show engine type for models when choosing models for transcription.

Documentation updates:

• Added documentation regarding starting a GCP account, billing, APIs and quotas setup.
• Cleaning up of existing documentation.

v1.0.3

• Fixed logging error caused by a Kaldi training stage failing.

v1.0.1

Added

• Resampling utility

Changed

• Reformated all Python code using Black formatter
• Clean-up of Dockerfile
• Caching of port-audio for Kaldi build
• Clean-up protobuf install in Dockerfile

v1.0.0

After several years of development Elpis has hit 1.0.0. There are still some rough edges but recent UI work by Ben and Matt have made using the platform a much more friendly experience.

Thank you to everyone who has contributed to bringing Elpis this far.

The Elpis Dev Team

v0.97.2

ADDED

• Delete buttons for datasets, pron dicts and models
• Minor bug fixes

v0.97.1

Changed

• 44.1kHz sample rate to 16kHz

v0.97.0

Added

• Huggingface Transformers wav2vec2 engine

Changed

• Fixed CLI examples

Removed

• ESPnet engine

v0.96.12

Automatic builds on releases