anggars/neural-mathrock

Neural Math Rock

Dataset Description

The Neural Math Rock dataset is a specialized, high-density audio collection built for deep learning applications in Music Information Retrieval (MIR), specifically focusing on the Math Rock and Midwest Emo subgenres. This dataset contains 4,000 embedded audio samples paired with detailed, multi-dimensional metadata.

It is designed to facilitate advanced audio classification tasks, including emotion recognition, vibe categorization, intensity mapping, and tempo analysis using multimodal deep learning architectures (e.g., WavLM, XLM-RoBERTa).

Dataset Specifications

• Total Samples: 4,000 audio tracks/chunks.
• Audio Profile: Embedded .wav binaries, optimized for feature extraction.
• Total Size: ~181 GB distributed across 400 Parquet shards to ensure optimal streaming and stable memory allocation during dataloading.

Features and Schema

The dataset is strictly structured and natively cast into Hugging Face's Audio feature class. Each record contains the following schema:

• artist (String): The name of the band or artist.
• song (String): The title of the track.
• file_name (String): The source filename.
• audio (Audio): The actual binary audio data, decodable on the fly.
• emotion (String): Primary emotional resonance of the track (e.g., optimism, sadness, admiration, realization, gratitude).
• vibe (String): Stylistic or atmospheric categorization.
• intensity (Float): Calculated energy or acoustic intensity of the track.
• tempo (Float): The detected tempo (BPM).

Usage

Due to the substantial size of the embedded audio files (158 GB), it is highly recommended to load this dataset using the streaming=True parameter.

from datasets import load_dataset

# Load the dataset in streaming mode
dataset = load_dataset("anggars/neural-mathrock", streaming=True)

# Fetch the first record
sample = next(iter(dataset['train']))

print(f"Artist: {sample['artist']}")
print(f"Song: {sample['song']}")
print(f"Emotion: {sample['emotion']}")
print(f"Tempo: {sample['tempo']} BPM")

# Access the raw audio array and sampling rate
audio_data = sample['audio']['array']
sampling_rate = sample['audio']['sampling_rate']