shaneweisz/swahili-keyword-spotting

Detecting the presence or absence of a particular word or phrase in a stream of Swahili audio # Swahili Keyword Spotting The aim of this project is to write a simple Keyword Spotting (KWS) system for Swahili. This involves detection of a particular word, or phrase, in a stream of Swahili audio. The KWS system will be based on the 1-best output with posterior scores (derived from confusion networks) from a speech recognition system. We will also investigate score-normalisation and system combination. ## Prerequisites * Python version >= 3.9 * Add appropriate folders to the `PYTHONPATH` environment variable to allow for relative imports by running: `source scripts/set-pythonpath.sh` ## Usage ### Running KWS on a single CTM file To run KWS on a CTM file, place the CTM file (e.g. `example.ctm`) in the `lib/ctms` folder and the appropriate queries XML file in the `lib/queries` folder, then run: ``` python3.9 src/kws/main.py --ctm example.ctm --queries queries.xml --output example-output.xml ``` This will output the XML hits output file to `outputs/example-output.xml` ### Running KWS on a chosen set of predefined CTM files Run: ``` scripts/run-kws-on-all-ctms.sh ``` Look closer at the script to see which systems this entails. This will output the XML hits output files to the `outputs` folder. ### Scoring the KWS output and getting the MTWV ``` scripts/score.sh output/example.xml scoring scripts/termselect.sh lib/queries/ivoov.map output/example.xml scoring [all|iv|oov] ``` ### Morphological Decomposition Decomposing `queries-word.xml` into `queries-morph.xml` ``` python3.9 src/morph-decomposition/queries_to_morphs.py ``` Decomposing `onebest-word.ctm` into `onebest-word-to-morphs.ctm` ``` python3.9 src/morph-decomposition/ctm_to_morphs.py ``` ### Running the unit tests: ``` python3.9 -m unittest discover -v -s ./src -p "*test*.py" ```