Elpis is a tool which allows language workers with minimal computational experience to build their own speech models for use in automatically transcribing audio. It relies on the Kaldi automatic speech recognition (ASR) library. Kaldi is notorious for being difficult to build, use and navigate - even for trained computer scientists. The goal of Elpis is to expose the power of Kaldi to linguists and language workers by abstracting away much of the needless technical complexity.
Elpis grew out of the kaldi_helpers project, a collection of Python and shell scripts designed to prepare data for use with Kaldi and convert between the various time-aligned transcription formats that linguists work with. These scripts now make up the core of Elpis' input preparation and output conversion engines.
Elpis uses Docker, specifically an Ubuntu Linux image, to install Kaldi and its (many) dependencies in an Ubuntu linux environment. Elpis itself consists of wrappers around Kaldi file system objects, a Flask server, and a web-based graphical user interface written in React (maintained in a separate repository here).
Please check the wiki pages for a step-by-step guide to using Elpis on your data.
If you want to play around with Elpis' internals, or contribute to development we recommend
cloning this repository and building the Docker image, but without the Docker ENTRYPOINT (you can simply comment it
out or pass the argument --entrypoint="bin/bash" when starting up the Docker image).
All of the required dependencies are included in the Docker image provided, including Kaldi, Python 3 etc.
Elpis is the Greek goddess of hope. For us Elpis represents our hope that one day everyone will have access to the power of ASR without having to interact directly with the Kaldi codebase. We've also backronymed it to stand for "Endangered Language Pipeline and Inference System," as much of our motivation for this project derives from our desire to assist in documentation and revitalisation efforts for the world's many endangered languages - including many in Australia, where most of Elpis' development has occurred.
This software is the product of academic research funded by the Australian Research Council Centre of Excellence for the Dynamics of Language. If you use the software in an academic setting, please be sure to cite it appropriately as follows:
Foley, B., Arnold, J., Coto-Solano, R., Durantin, G., Ellison, T. M., van Esch, D., Heath, S., Kratochvíl, F., Maxwell-Smith, Z., Nash, D., Olsson, O., Richards, M., San, N., Stoakes, H., Thieberger, N. & Wiles, J. (2018). Building Speech Recognition Systems for Language Documentation: The CoEDL Endangered Language Pipeline and Inference System (Elpis). In S. S. Agrawal (Ed.), The 6th Intl. Workshop on Spoken Language Technologies for Under-Resourced Languages (SLTU) (pp. 200–204). Available on https://www.isca-speech.org/archive/SLTU_2018/pdfs/Ben.pdf.
