Free and open-source ser_model_fixed_label Speech Emotion Recognition Model - Accuracy on the evaluation set reaches 83.67%

Ser Model Fixed Label

Developed by aherzberg

A speech emotion recognition model fine-tuned based on facebook/wav2vec2-base, achieving an accuracy of 83.67% on the evaluation set

Downloads 18

Release Time : 2/28/2023

Model Overview

This model is a speech emotion recognition model based on the wav2vec2 architecture, used to identify emotion categories from speech

High Accuracy

Achieves 83.67% accuracy on the evaluation set

Based on wav2vec2 Architecture

Uses facebook's wav2vec2-base as the base model

End-to-End Training

Directly processes raw audio input without complex feature engineering

Speech Emotion Recognition

Audio Classification

Sentiment Analysis

Customer Service Call Analysis

Analyze customer emotional states in service calls

Can recognize 83.67% of emotion categories

Mental Health Assessment

Assess user emotional states through speech analysis

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.7645	0.96	18	1.5899	0.4333
1.5148	1.96	36	1.4152	0.4433
1.3042	2.96	54	1.1857	0.5767
1.1184	3.96	72	1.0508	0.62
0.9588	4.96	90	0.9329	0.7
0.9789	5.96	108	0.8638	0.74
0.7835	6.96	126	0.7730	0.8133
0.7259	7.96	144	0.7355	0.83
0.6783	8.96	162	0.7190	0.8333
0.6644	9.96	180	0.7010	0.8367

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base