Hubert-base-ls960-finetuned-gtzan Open-source Audio Classification Model

Hubert Base Ls960 Finetuned Gtzan

Developed by c72599

An audio classification model based on the HuBERT architecture, fine-tuned on the GTZAN music genre classification dataset with an accuracy of 88%

Audio Classification

Transformers

Open Source License:Apache-2.0 #Music Classification #High Accuracy Audio Recognition #HuBERT Fine-tuning

Downloads 15

Release Time : 7/29/2023

Model Overview

This model is an audio classification model fine-tuned on the GTZAN music dataset, based on the facebook/hubert-base-ls960 pre-trained model, specifically designed for music genre classification tasks.

Model Features

High Accuracy

Achieves 88% classification accuracy on the GTZAN test set

Pre-trained + Fine-tuning Architecture

Based on the large-scale pre-trained HuBERT model, fine-tuned for specific tasks

Efficient Training

Optimizes the training process using techniques like gradient accumulation

Model Capabilities

Music Genre Classification

Audio Feature Extraction

Music Content Analysis

Use Cases

Music Analysis

Automatic Music Genre Classification

Classifies music clips into genres (e.g., rock, jazz, classical)

88% Accuracy

Music Recommendation System

Serves as a feature extraction component for music recommendation systems

Audio Processing

Audio Content Analysis

Analyzes audio content features

🚀 hubert-base-ls960-finetuned-gtzan

This model is a fine - tuned version of facebook/hubert-base-ls960 on the GTZAN dataset. It can be used for audio classification tasks and achieves an accuracy of 0.88 on the evaluation set.

📚 Documentation

Model Information

Property	Details
Model Type	Fine - tuned version of facebook/hubert-base-ls960
Training Data	marsyas/gtzan
Metrics	Accuracy

This model achieves the following results on the evaluation set:

Loss: 0.6645
Accuracy: 0.88

Training and Evaluation

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 2
eval_batch_size: 2
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 16
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
2.2685	1.0	56	2.2069	0.44
2.0208	1.99	112	1.8352	0.46
1.7603	2.99	168	1.5275	0.49
1.4843	4.0	225	1.4296	0.52
1.347	5.0	281	1.2222	0.52
1.2364	5.99	337	1.1477	0.62
1.2082	6.99	393	1.0181	0.67
0.9861	8.0	450	0.9598	0.71
0.752	9.0	506	0.7499	0.77
1.006	9.99	562	0.8190	0.79
0.6725	10.99	618	0.8798	0.75
0.7457	12.0	675	0.6276	0.81
0.4605	13.0	731	0.6086	0.85
0.5751	13.99	787	0.6894	0.75
0.4886	14.99	843	0.6109	0.83
0.2429	16.0	900	0.6076	0.85
0.3084	17.0	956	0.4646	0.86
0.3762	17.99	1012	0.8349	0.81
0.2897	18.99	1068	0.4509	0.89
0.1296	20.0	1125	0.6791	0.86
0.1291	21.0	1181	0.6466	0.85
0.3784	21.99	1237	0.6272	0.88
0.1156	22.99	1293	0.7916	0.85
0.2093	24.0	1350	0.6536	0.85
0.2167	25.0	1406	0.7050	0.87
0.1095	25.99	1462	0.6128	0.88
0.1004	26.99	1518	0.6092	0.89
0.0897	28.0	1575	0.6730	0.88
0.083	29.0	1631	0.6396	0.89
0.0343	29.87	1680	0.6645	0.88

Framework versions

Transformers 4.32.0.dev0
Pytorch 1.13.1+cu116
Datasets 2.14.1
Tokenizers 0.13.3

📄 License

This model is licensed under the Apache 2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご