Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.
Date is year-month-day format.

[1.0.6] - 2022-11-03

Changes:

Fix model download bug due to change in Flask

[1.0.5] - 2022-10-21

Changes:

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

[1.0.4] - 2022-10-17

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.

[1.0.3] - 2022-09-21

Changed

Fixed logging error caused by a Kaldi training stage failing.

[1.0.2] - 2022-09-20

Changed

Fixed white screen of death when uploading a single wav to datasets
Fixed HFT failing on first transcription due to it expecting audio.wav name
Fixed yarn watch for devs

[1.0.1] - 2022-06-28

Added

Resampling utility

Changed

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

[0.97.2] - 2021-12-23

Added

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

[0.97.1] - 2021-12-08

Changed

44.1kHz sample rate to 16kHz

[0.97.0] - 2021-12-05

Added

Huggingface Transformers wav2vec2 engine

Changed

Fixed CLI examples

Removed

ESPnet engine

[0.96.12] - 2021-11-10

Added

Automated docker builds on releases (vX.Y.Z) and pushes to master (latest)

[0.96.11] - 2021-11-09

Changed

Change ports from 5000 to 5001 to avoid conflict with greedy MacOS Airplay speaker thing

[0.96.10] - 2021-10-28

Changed

Show Kaldi hypothesis confidence on the transcription page

[0.96.9] - 2021-08-27

Changed

Updated pympi, praatio, llvm, librosa, numba versions

[0.96.7] - 2021-03-23

Added

Front end and dynamic i18n support

[0.96.6] - 2021-03-6

Changed

Minor gui tweaks

[0.96.5] - 2021-03-5

Changed

New flow for GUI welcome and engine select
Start implementing feedback from UX study
Show train logs in GUI. Kaldi is split into stages, ESPnet is single stage.
Add DEV_MODE setting for GUI
Pin ESPnet to CoEDL fork of persephone-tools/espnet

[0.96.4] - 2021-03-4

Changed

Changed to use pyenv for Python version management

[0.96.3] - 2021-01-26

Changed

Voice Activity Detection
Bring GUI Dockerfile into line with with CPU version

[0.96.2] - 2021-01-07

Changed

Modify Elpis to use Poetry for dependency management

[0.96.1] - 2021-01-05

Changed

Dockerfile to use Python 3.8 as default for Elpis

[0.96.0] - 2020-12-29

Changed

Monorepo: brought https://github.com/coedl/elpis-gui into this repo under elpis/gui
Significant restructure: moved gpu and examples under /elpis
Updated Docker image to an Ubuntu 20.04 base image

[0.95.2] - 2020-12-08

Changed

Dockerfile to use Python 3.7 as default for Elpis

[0.95.1] - 2020-09-23

Changed

Rearranged Dockerfile to group like things and install ESPne earlier

[0.95.0] - 2020-09-20

Added

ESPnet

Changed

Fixed changelog date for docs

[0.94.6] - 2020-09-07

Added

Docs

Changed

Separated log files into individual logs for each stage while processing, and build complete log when done
Fixed annotations not resetting when changing tier settings
Moved CLI example into a Kaldi dir
Fixed Kaldi Elan CLI example, now links dataset and pronunciation dictionary independently