| ---
|
| license: other
|
| task_categories:
|
| - audio-classification
|
| tags:
|
| - audio
|
| - road-sounds
|
| - sound-event-detection
|
| - environmental-sound
|
| size_categories:
|
| - n<1K
|
| ---
|
|
|
| # MIVIA Road Events Dataset
|
|
|
| ## Dataset Summary
|
|
|
| MIVIA_ROAD_DB1 is an annotated audio dataset for **road acoustic event detection**.
|
| 57 scene recordings with precise timestamped annotations for two classes of road
|
| sound events. Originally released by the **MIVIA Lab, University of Salerno**.
|
|
|
| ## Dataset Viewer
|
|
|
| The viewer above shows one scene audio file per row. Use `MIVIA_events.csv` for
|
| the full per-event annotation table (400 events total).
|
|
|
| ## Statistics
|
|
|
| | | |
|
| |---|---|
|
| | Scene recordings | 57 |
|
| | Total events | 400 |
|
| | Class 2 events | 200 |
|
| | Class 3 events | 200 |
|
| | Audio format | WAV, 32000 Hz, mono |
|
| | Avg duration | ~65–70 seconds |
|
|
|
| ## Files
|
|
|
| | File | Description |
|
| |---|---|
|
| | `data/*.wav` | 57 scene audio recordings |
|
| | `metadata.csv` | Scene-level metadata (powers the viewer) |
|
| | `MIVIA_events.csv` | Full event table (400 rows, timestamps per event) |
|
| | `MIVIA_events.json` | Same data in JSON format |
|
|
|
| ## Columns in MIVIA_events.csv
|
|
|
| | Column | Description |
|
| |---|---|
|
| | `xml_file` | Source annotation XML |
|
| | `scene_audio` | Mixed scene WAV file |
|
| | `event_wav` | Isolated event sound file |
|
| | `class_id` | 2 or 3 |
|
| | `class_name` | Event class name |
|
| | `start_sec` | Event start time (seconds) |
|
| | `end_sec` | Event end time (seconds) |
|
| | `bg_class_name` | Background sound class |
|
| | `bg_pathname` | Background audio file |
|
|
|
| ## Usage
|
|
|
| ```python
|
| import pandas as pd
|
| from huggingface_hub import hf_hub_download
|
|
|
| # Load full event annotations
|
| csv_path = hf_hub_download(
|
| repo_id="Titung/MIVIA",
|
| filename="MIVIA_events.csv",
|
| repo_type="dataset"
|
| )
|
| df = pd.read_csv(csv_path)
|
| print(df.head())
|
| print(df["class_id"].value_counts())
|
| ```
|
|
|
| ```python
|
| # Load a scene audio file
|
| from huggingface_hub import hf_hub_download
|
| import librosa
|
|
|
| wav_path = hf_hub_download(
|
| repo_id="Titung/MIVIA",
|
| filename="data/00001_1.wav",
|
| repo_type="dataset"
|
| )
|
| y, sr = librosa.load(wav_path, sr=None)
|
|
|
| # Slice first event
|
| row = df.iloc[0]
|
| event_audio = y[int(row.start_sec * sr) : int(row.end_sec * sr)]
|
| print(f"Event duration: {len(event_audio)/sr:.2f}s Class: {row.class_id}")
|
| ```
|
|
|
| ## Citation
|
|
|
| > Foggia, P., Petkov, N., Saggese, A., Strisciuglio, N., & Vento, M. (2015).
|
| > *Reliable detection of audio events in highly noisy environments.*
|
| > Pattern Recognition Letters. Elsevier.
|
|
|
| MIVIA Lab, University of Salerno — http://mivia.unisa.it
|
|
|
