| USAGE.txt - Procedural Engine Sounds Dataset |
| ========================================== |
|
|
| QUICK START GUIDE |
| ----------------- |
|
|
| This dataset contains 4-channel WAV audio files with engine sounds and embedded RPM/torque data. |
|
|
| FILE STRUCTURE: |
| - audio/[A-H]_*_set/: WAV files organized in 8 sets |
| - metadata/: JSON summaries and CSV statistics for each set |
|
|
| AUDIO CHANNEL LAYOUT: |
| - Channel 1-2: Stereo engine sound audio |
| - Channel 3: RPM values (multiply by 10,000 for actual RPM) |
| - Channel 4: Torque values (multiply by 1,000 for actual Newton-meters) |
|
|
| BASIC USAGE EXAMPLES: |
|
|
| 1. PYTHON (with soundfile): |
| ```python |
| import soundfile as sf |
| import numpy as np |
| |
| # Load a 4-channel audio file |
| audio, sr = sf.read('audio/A_full_set/engine_001.wav') |
| |
| # Extract channels |
| engine_left = audio[:, 0] # Left engine audio |
| engine_right = audio[:, 1] # Right engine audio |
| rpm_signal = audio[:, 2] * 10000 # RPM values |
| torque_signal = audio[:, 3] * 1000 # Torque in Nm |
| ``` |
|
|
| 2. PYTHON (with librosa): |
| ```python |
| import librosa |
| |
| # Load specific channels |
| engine_audio, sr = librosa.load('audio/A_full_set/engine_001.wav', |
| sr=48000, mono=False) |
| # Result shape: (4, samples) - channels x samples |
| ``` |
|
|
| 3. MATLAB: |
| ```matlab |
| [audio, fs] = audioread('audio/A_full_set/engine_001.wav'); |
| |
| engine_left = audio(:,1); |
| engine_right = audio(:,2); |
| rpm = audio(:,3) * 10000; |
| torque = audio(:,4) * 1000; |
| ``` |
|
|
| METADATA ACCESS: |
|
|
| 1. Load set summary (JSON): |
| ```python |
| import json |
| with open('metadata/A_full_set_summary.json', 'r') as f: |
| summary = json.load(f) |
| print(f"Set A contains {summary['num_files']} files") |
| ``` |
|
|
| 2. Load file statistics (CSV): |
| ```python |
| import pandas as pd |
| stats = pd.read_csv('metadata/A_full_set_stats.csv') |
| print(stats.head()) # View first few files' stats |
| ``` |
|
|
| DATASET ORGANIZATION: |
| - Full Sets (A,B,C,D): ~767 files each, ~2.46 hours each |
| - Large Sets (E,F,G,H): ~717 files each, ~2.30 hours each |
| - Total: 5,935 files, ~19 hours, ~24.5 GB |
|
|
| TECHNICAL SPECS: |
| - Sample Rate: 48 kHz |
| - Format: WAV (uncompressed) |
| - Channels: 4 (quad) |
| - Total Duration: ~19.01 hours |
| - Total Size: ~24.47 GB |
|
|
| COMMON WORKFLOWS: |
|
|
| 1. AUDIO ANALYSIS: |
| - Load engine audio (channels 1-2) |
| - Analyze spectral content, harmonics |
| - Correlate with RPM/torque data (channels 3-4) |
|
|
| 2. MACHINE LEARNING: |
| - Use engine audio as input features |
| - Use RPM/torque as target labels |
| - Train regression or classification models |
|
|
| 3. AUDIO SYNTHESIS: |
| - Study relationship between RPM/torque and audio features |
| - Train generative models conditioned on engine parameters |
| - Use for data augmentation in automotive applications |
|
|
| REQUIREMENTS: |
| - Audio processing software supporting multi-channel WAV |
| - ~25 GB free disk space |
| - Python/MATLAB/R with audio processing libraries (recommended) |
|
|
| TROUBLESHOOTING: |
| - If files won't load: Check multi-channel WAV support in your software |
| - If values seem wrong: Remember to apply scaling (RPM×10000, Torque×1000) |
| - For large dataset: Consider loading files individually rather than all at once |
|
|
| LICENSE: CC BY-NC 4.0 (Attribution-NonCommercial) |
| CONTACT: doerflerrobin@gmail.com |
|
|
| For detailed documentation, see README.txt |
