D

Duck And Cover Genre Encoder

Developed by mnne
A music genre encoder fine-tuned based on BERT-mini, capable of embedding music genres into a 256-dimensional space with an accuracy rate exceeding 98%.
Downloads 16
Release Time : 4/2/2022

Model Overview

This model is part of the duck_and_cover project, used to generate album covers based on music genres. It fine-tunes the BERT-mini model to encode genre information and trains a linear layer for genre prediction.

Model Features

High Accuracy Genre Encoding
Achieves 98.29% test accuracy on 3,452 labels, effectively handling complex genre combinations.
Semantic-aware Encoding
Utilizes BERT tokenizer to capture semantic relationships between genres (e.g., the connection between 'hard rock' and 'rock').
Large-scale Training Data
Trained on genre data from 466,045 real albums, covering a wide range of music types.

Model Capabilities

Music Genre Embedding
Multi-label Classification
Semantic Relationship Capture

Use Cases

Music Recommendation Systems
Genre Similarity Calculation
Discovers albums with similar styles by comparing genre embedding vectors.
Enhances the relevance of music recommendations.
Music Information Retrieval
Smart Genre Filtering
Converts user queries into genre embeddings for precise matching.
Improves the accuracy of music library searches.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase