Introduction

This is the source code of the paper A Neural-network-based Approach to Identifying Speakers in Novels presented on Interspeech 2021. It demonstrates the method to detect the speakers of the utterances in the novel given a fix range of contexts. Speaker-alternation-based revision is implemented in the evaluation script. A copy of the dataset on https://github.com/YueChenkkk/Chinese-Dataset-Speaker-Identification is included in ./data directory.

Usage

For training: bash run_train.sh
For evaluation: python evaluate.py

Requirements

python==3.6
jieba==0.42.1
numpy==1.19.1
torch==1.2.0
tensorflow==2.0.0
transformers==4.6.1
fastprogress==1.0.0