lingsoft/memad-lidbox repository overview

⁠ELG API for MeMAD-lidbox

ELG compatible⁠ Flask based REST API for the spoken language identification tool created in MeMAD⁠ project.

⁠Deploying the service

docker run -d -p <port>:8000 --init lingsoft/memad-lidbox:<tagname>

To prevent the critical worker timeout error you may need to increase environment variable TIMEOUT. The default value is now only 60 seconds. Add --env TIMEOUT=xxx to call. You can also set the number of the workers in the same way.

⁠Usage

See instructions from GitHub⁠. The API can identify these languages: fi, sv, fr, de, en, and x-nolang (denotes no language detected).

The pipeline works in two scenarios:

• if there is only an audio file in the request, the API splits the input audio into 2 seconds chunks and predicts corresponding spoken languages.
• if there is an audio file and corresponding annotation/diarization json, defining the starting and ending times of speech fragments and their optional language labels, the API returns prediction results for the corresponding fragments.

⁠Authors

⁠Original authors

Original pipeline was created in the project memad-lid-pipeline⁠, the author of the pipeline is LimeCraft⁠. The tool lidbox⁠ is developed by Matias Lindgren with MIT license.

⁠ELG API integration

The ELG API was developed by Lingsoft⁠ as part of the Connecting Europe Facility (CEF) project Microservices at your service: bridging the gap between NLP research and industry⁠

⁠License

MIT