vancenceho/youtube-spotify-audio-features

---
license: cdla-sharing-1.0
language:
- en
tags:
- music
pretty_name: Spotify-YouTube Audio Features
size_categories:
- 10K<n<100K
---

# Spotify–YouTube Audio Features

Tabular **librosa** audio features for tracks aligned with the Spotify / YouTube pipeline in the *viral-content-predictor* project. Each row is one Spotify `track_id` matched to a downloaded YouTube audio clip; features are aggregated statistics (mean / std) computed on the decoded waveform.

## Files

| File | Description |
|------|-------------|
| `audio_features.csv` | One row per track: `track_id`, 89 derived feature dimensions (means/stds), `extraction_success`, `error_message`. |

## Data instance

- **Format:** CSV (header row).
- **Rows:** on the order of **tens of thousands** of tracks (exact count may change as the extraction job is updated).
- **Key column:** `track_id` — Spotify track identifier, join key to other project tables.
- **Label / target:** not included; this file is **features only**.

## Feature groups (89 dimensions)

Features are extracted with **librosa** from MP3s under the project’s YouTube-audio download step (`02b_youtube_audio_feature_extraction.ipynb`). Groups include:

- Tempo (1)
- RMS energy (2)
- Zero crossing rate (2)
- Spectral centroid, rolloff, bandwidth (6)
- Spectral contrast — 7 bands (14)
- MFCC — 13 coefficients (26)
- Chroma — 12 pitch classes (24)
- Tonnetz (12)
- Onset strength (2)

Most groups contribute **mean** and **std** columns; see the CSV header for exact names.

Additional columns:

- **`extraction_success`** — boolean-like flag indicating whether feature extraction completed for that file.
- **`error_message`** — empty or diagnostic text when extraction failed.

## Provenance

- **Audio source:** user-downloaded YouTube audio (`.mp3`) matched to Spotify metadata elsewhere in the pipeline.
- **Processing:** Python, **librosa**; optional multiprocessing for throughput.
- **This artifact:** not the raw audio; only numeric summaries suitable for modeling and joins.

## Usage

```python
import pandas as pd

df = pd.read_csv("audio_features.csv")
# Join on track_id with Spotify / YouTube tables in your project
```

## Limitations

- Coverage is limited to tracks with a successful YouTube match and a readable audio file.
- Feature definitions follow librosa defaults; hyperparameters (e.g. hop length, FFT size) are fixed in the extraction notebook.

## License

This dataset card specifies **CDLA-Sharing-1.0**. Ensure your redistribution of the CSV and any derived data complies with that license and with the licenses of the underlying **Spotify** and **YouTube** data you used to build the file.

---
license: cdla-sharing-1.0
language:
- en
tags:
- music
pretty_name: Spotify-YouTube音频特征
size_categories:
- 10K<n<100K
---

# Spotify-YouTube音频特征

本数据集为**viral-content-predictor**项目中适配Spotify/YouTube流水线的曲目提供基于librosa（librosa）的表格型音频特征。每一行对应一个与下载的YouTube音频片段匹配的Spotify `track_id`；特征为对解码波形计算得到的聚合统计量（均值/标准差）。

## 文件

| 文件 | 描述 |
|------|-------------|
| `audio_features.csv` | 每个曲目一行，包含`track_id`、89个衍生特征维度（均值/标准差）、`extraction_success`、`error_message`。 |

## 数据实例

- **格式：** CSV（带表头行）。
- **行数：** 约数万条曲目（准确数量会随特征提取任务更新而变动）。
- **关键列：** `track_id` — Spotify曲目标识符，用于与项目内其他表格进行连接。
- **标签/目标变量：** 未包含；本文件仅包含特征数据。

## 特征组（共89维）

特征通过**librosa**从项目YouTube音频下载步骤（`02b_youtube_audio_feature_extraction.ipynb`）生成的MP3文件中提取，特征组包括：
- 节奏（1维）
- RMS能量（2维）
- 过零率（2维）
- 频谱质心、频谱滚降、频谱带宽（6维）
- 频谱对比度——7个频带（14维）
- Mel频率倒谱系数（MFCC）——13个系数（26维）
- 色度特征——12个音级（24维）
- Tonnetz（12维）
- 起音强度（2维）

大部分特征组会生成**均值**与**标准差**列，具体列名可查看CSV文件表头。

### 附加列
- **`extraction_success`**：类布尔标志，用于指示该文件的特征提取是否完成。
- **`error_message`**：提取失败时为空或包含诊断文本。

## 数据来源
- **音频来源：** 用户下载的YouTube音频（`.mp3`），已与流水线其他环节的Spotify元数据完成匹配。
- **处理流程：** 基于Python与**librosa**实现，支持可选的多进程以提升处理吞吐量。
- **本数据集产物：** 非原始音频，仅包含适用于建模与表格连接的数值型汇总数据。

## 使用示例
python
import pandas as pd

df = pd.read_csv("audio_features.csv")
# 基于`track_id`与项目内其他Spotify/YouTube表格进行关联

## 局限性
- 数据覆盖范围受限：仅包含成功匹配到YouTube音频且音频文件可正常读取的曲目。
- 特征定义遵循librosa的默认参数；超参数（如跳变长度hop length、快速傅里叶变换FFT大小）已在特征提取笔记本中固定。

## 许可证
本数据集卡片遵循**CDLA-Sharing-1.0**许可证。请确保您对该CSV文件及其衍生数据的再分发符合该许可证要求，同时也需符合构建本文件所使用的**Spotify**与**YouTube**底层数据的许可证条款。