# Gender Recognition — Data

来源：https://www.kaggle.com/competitions/gender-recognition-upc-20252026/data

采集时间：2026-10-24T17:16:59Z

This dataset is a cleaned and processed version of raw audio files for gender classification. The features were extracted from .wav audio recordings collected in a quiet room with no background noise. The data contains no null or duplicate values, ensuring a high-quality starting point for analysis and modeling.

## Features:
The dataset includes the following extracted audio features:

mean_spectral_centroid: The average spectral centroid, representing the "center of mass" of the spectrum, indicating brightness.

std_spectral_centroid: The standard deviation of the spectral centroid, measuring variability in brightness.

mean_spectral_bandwidth: The average width of the spectrum, reflecting how spread out the frequencies are.

std_spectral_bandwidth: The standard deviation of spectral bandwidth, indicating variability in frequency spread.

mean_spectral_contrast: The average difference between peaks and valleys in the spectrum, indicating tonal contrast.

mean_spectral_flatness: The average flatness of the spectrum, measuring the noisiness of the signal.

mean_spectral_rolloff: The average frequency below which a specified percentage of the spectral energy resides, indicating sharpness.

zero_crossing_rate: The rate at which the signal crosses the zero amplitude axis, representing noisiness or percussiveness.

rms_energy: The root mean square energy of the signal, reflecting its loudness.

mean_pitch: The average pitch frequency of the audio.

min_pitch: The minimum pitch frequency.

max_pitch: The maximum pitch frequency.

std_pitch: The standard deviation of pitch frequency, measuring variability in pitch.

spectral_skew: The skewness of the spectral distribution, indicating asymmetry.

spectral_kurtosis: The kurtosis of the spectral distribution, indicating the peakiness of the spectrum.

energy_entropy: The entropy of the signal energy, representing its randomness.

log_energy: The logarithmic energy of the signal, a compressed representation of energy.

mfcc_1_mean to mfcc_13_mean: The mean of the first 13 Mel Frequency Cepstral Coefficients (MFCCs), representing the timbral characteristics of the audio.

mfcc_1_std to mfcc_13_std: The standard deviation of the first 13 MFCCs, indicating variability in timbral features.
label: The target variable indicating the gender male(1) or female(0).

## Key Information:

Clean Data: The dataset has been thoroughly cleaned and contains no null or duplicate values.

Unscaled: The features are not scaled, allowing users to apply their preferred scaling or normalization techniques.

Feature Extraction: The function used for feature extraction is available in the notebook in the Code section.

High Performance: The data achieved 95%+ accuracy using machine learning models such as Random Forest, Extra Trees, and K-Nearest Neighbors (KNN). It also performed exceptionally well with neural networks.

## Recommendations:

Feature Selection: Avoid using all features in modeling to prevent overfitting. Instead, perform feature selection and choose the most impactful features based on your analysis.

This processed dataset is a reliable and robust foundation for building high-performing models.