Implementation of the approaches used in the paper: "Topic identification for spontaneous speech: Enriching audio features with embedded linguistic information".
The models are implemented using the SpeechBrain toolkit and the recipes are available in the subdirectories.
An overview of the explored Topic ID systems is given in the figure below:
To run the experiments, you will need the following dependencies:
To execute a recipe, you need to run the train and hyperparams files, for example:
python ctc_aed_train.py ctc_aed_hyperparams.yaml
The pre-trained models can be obtained from: https://zenodo.org/records/10158851
Cite the paper:
Porjazovski, D., Grósz, T., & Kurimo, M. (2023). Topic Identification for Spontaneous Speech: Enriching Audio Features with Embedded Linguistic Information. In *2023 31st European Signal Processing Conference (EUSIPCO)* (pp. 396–400). IEEE.