Annotated dataset of simulated voiding sound for urine flow estimation

# Annotated dataset of simulated voiding sound for urine flow estimation

## Overview

This repository contains a dataset of synthetic urination audio signals generated under controlled conditions. The dataset is intended for research in sound-based uroflowmetry (SU) and the development of machine learning models for voiding flow estimation. The audio samples were generated using a high-precision peristaltic pump to simulate flow rates between 1 and 50 ml/s and were recorded with three different microphone devices.

## Experimental Setup

The audio recordings were conducted in a bathroom setup where the synthetic urine stream generated by the peristaltic pump was directed into a standard ceramic toilet containing a fixed volume of water at the bottom, ensuring that the sound was produced by the interaction of the liquid stream with the water surface. This configuration mimics realistic voiding conditions and allows the captured audio to resemble actual urination events.

## Dataset Structure

Each audio file is a 60-second segment labeled with its corresponding flow rate (in ml/s). The naming convention is:

[device]_f_[flow]_[duration]s.wav

- `device`: UM (Ultramic384k), Phone (Mi A1), or Watch (Oppo Smartwatch)
- `flow`: flow rate from 1 to 50 ml/s
- `duration`: fixed to 60 seconds for all files

### Example:

um_f_20_60s.wav phone_f_45_60s.wav watch_f_10_60s.wav

## Silence Reference Recordings

In addition to the voiding audio files, each device folder includes a 30-second silence recording. These recordings were captured in the same environment and setup, but without any synthetic flow, allowing for baseline noise analysis. They serve as a reference to evaluate the background noise characteristics of each device and to support preprocessing techniques such as noise reduction or signal enhancement.

Filename format:

[device]_f_0_30s.wav

### Example:
- `um_f_0s.wav`
- `phone_f_0s.wav`
- `oppo_f_0s.wav`

## Purpose

The goal of this dataset is to provide a standardized audio repository for the development, training and validation of machine learning algorithms for voiding flow prediction. This enables researchers to:
- Benchmark different approaches on a common dataset
- Develop flow estimation models using synthetic audio before transferring them to real-world applications
- Explore the spectral and temporal structure of urination-related audio signals

## Flow Generation

- Pump Used: L600-1F precision peristaltic pump
- Flow Range: 1–50 ml/s (based on ICS-reported ranges for male uroflowmetry)
- Calibration: Pump flows were validated using a graduated cylinder
- Noise Isolation: The pump was placed in a separate room (via 15m silicone tubing) to eliminate pump noise from recordings

## Recording Devices

| Device | Sampling Rate | Frequency Range | Description |
|---------|----------------|------------------|--------------------------------------|
| UM | 192 kHz | 0–96 kHz | High-quality ultrasonic microphone |
| Phone | 48 kHz | 0–24 kHz | Android smartphone (Mi A1) |
| Watch | 44.1 kHz | 0–22.05 kHz | Oppo Smartwatch with built-in mic |

Each recording was carried out using a custom mobile or desktop app with preset parameters.

## Recording Environment

- Recordings were made in a bathroom with a standard ceramic toilet containing water at the bottom.
- The nozzle height varied between 73–86 cm depending on flow rate to ensure consistent water impact.
- Microphone heights:
- UM: 84 cm
- Phone: 95 cm
- Watch: 86 cm (simulating wrist height)

## Data Collection Protocol

1. Pump activated with flow set from 1 to 50 ml/s.
2. Audio recorded simultaneously with UM, Phone and Watch for 80 seconds.
3. Initial 15 seconds and final 5 seconds trimmed to retain 60 seconds of steady-state urination sound.

## Citation

If you use this dataset in your work, please cite the associated paper:

> M. L. Alvarez et al., “Annotated dataset of simulated voiding sound for urine flow estimation”, 2025. (Pending publication)

## License

This dataset is made available for research purposes under a CC BY license.

## Contact

For questions, please contact:

Marcos Lazaro Alvarez
Faculty of Engineering, University of Deusto
[e-mail]