MULTILINGUAL SPEECH EMOTION RECOGNITION WITH MULTI-GATING MECHANISM AND NEURAL ARCHITECTURE SEARCH

Prerequisite

1. Install packages

use the following command to install packages:

pip install -r requirements.txt

The cuda version is 11.1.

2. Download Dataset and Features

Wav files: https://drive.google.com/drive/folders/12V4A78u_XOp7vXlUrF5vweXEF9nb4btJ?usp=sharing

Allosaurus features: https://drive.google.com/drive/folders/1YDjsl05ubArpvo4kk9yzXjnbexXgInHP?usp=sharing

GE2E features is in data/GE2E

MFCC features: https://drive.google.com/drive/folders/12aaC-n4o0qXSL53rRFcLWqx6ar1J4iUs?usp=sharing

Train the Model

Firstly, you need to change the paths in file model/SER_mmoe/training/utils.py and model/SER_mmoe/modeling/ser_model.pyto the path in your own environment.

The training process is very simple. Please run the following command:

cd model/
sh train.sh