numblilbug/diarization_khoekhoe

Repository for a course project on the topic 'Speaker Diarization on the data of Low-Resource Languages' # diarization_khoekhoe Repository for a course project on the topic: **"Speaker Diarization on the data of Low-Resource Languages"** ## Notebooks Overview - `creatingprotocol.ipynb`: Takes as input ELAN-annotated files with the `.eaf` extension (the files themselves and a list of their names in a CSV) and creates a database `database.yml`. The database contains links to three types of files: - `.rttm` — files with annotated segments of each audio - `.uem` — annotation boundaries - `.lst` — list of audio filenames - `finetuningandoptimizing.ipynb`: Takes `database.yml` as input and fine-tunes the model and optimizes parameters based on the available data. The code in this notebook is adapted from the tutorial: github.com - `correlations.ipynb`: Calculates correlations between DER (Diarization Error Rate) and parameters that may be significant, such as: - Total speech duration in the file - Spoken language - Number of speakers