Starstruk Logo MarkStarstruk

v0.9.1

HumanSignal/label-studiov0.9.1Feb 16, 2021by makseq

AI Summary

This release focuses on cloud storage improvements (adding Azure and Redis support, plus S3 endpoint configuration), introduces Automated Speech Recognition with Nvidia NeMo ML backend, and includes various bug fixes for export duplicates, multi-session uploads, and non-ASCII language support.

Key Highlights

  • Added Microsoft Azure and Redis as source/target storage options for tasks and completions
  • Introduced Automated Speech Recognition (ASR) with NVIDIA NeMo toolkit and new ASR_MANIFEST export format
  • Enabled S3 endpoint configuration via environment variable
  • Added support for non-ASCII languages like Persian
  • Fixed critical bugs: duplicate completions in exports, multi-session file uploads, and .conll format label issues

New Features

  • Microsoft Azure cloud storage support (source and target)
  • Redis cloud storage support (source and target)
  • Amazon S3 endpoint configuration via environment variable
  • Automated Speech Recognition ML backend using NVIDIA NeMo
  • ASR_MANIFEST export format
  • POST label config as XML using the API
  • Pre-labeling to make predictions for specific tasks by ID
  • Non-ASCII language support (e.g., Persian)
  • Renamed 'Whole Audio Transcription' template to 'Speech Transcription'
  • Improved timeseries data import flow
  • Enhanced multi-session mode experience

Full Release Notes

## Cloud storage improvements
- Set up Microsoft Azure as source or target storage for tasks and completions ([@keesschollaart81](https://github.com/keesschollaart81)). 
- Set up Redis as source or target storage for tasks and completions. ([@kasuteru](https://github.com/kasuteru))
- Added an option to set the Amazon AWS S3 endpoint with an environment variable when setting up cloud storage ([@herber523](https://github.com/herber523)).

See [Cloud Storages](https://labelstud.io/guide/storage.html).
## Automated Speech Recognition with Nvidia NeMo ML backend
- Added an example for Automatic Speech Recognition (ASR) machine learning backend using NVIDIA NeMo toolkit, including an ASR_MANIFEST export format.
- Updated the Label Studio Frontend for textarea labeling to make it easier to update transcription labels.
- Renamed the template "Whole Audio Transcription" to "Speech Transcription".
## Other improvements
- POST a label config as XML using the API ([@laynr](https://github.com/laynr)).
- Add a while loop to check for more busy ports ([@UtrechtUniversity](https://github.com/UtrechtUniversity)).
- Use pre-labeling to make predictions for specific tasks by ID. 
- No longer require tasks and completions to be ASCII-formatted, supporting non-ASCII languages like Persian ([#578](https://github.com/heartexlabs/label-studio/issues/578))
- Improved the workflow for importing data to make it easier to fix label configuration errors.
- Improved the timeseries data import flow to make it clearer how to import from text files. 
- Improved the multi-session mode experience.
## Notable bug fixes 
- Fixed an issue where exported data would have duplicate completions.
- Fixed a bug when uploading files in multi-session mode.
- Improved handling of timeseries and polygon labels.
- Fixed a bug where the .conll export format misses labels or assigns them to the subsequent token.