Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Machine listening in a neonatal intensive care unit

Modan Tailleur, Vincent Lostanlen, Jean-Philippe Rivière, Pierre Aumond

TL;DR

The feasibility of polyphonic machine listening in a hospital ward while guaranteeing privacy by design is demonstrated and a small-scale study in a neonatological intensive care unit confirms that the time series of detected events align with another modality of measurement.

Abstract

Oxygenators, alarm devices, and footsteps are some of the most common sound sources in a hospital. Detecting them has scientific value for environmental psychology but comes with challenges of its own: namely, privacy preservation and limited labeled data. In this paper, we address these two challenges via a combination of edge computing and cloud computing. For privacy preservation, we have designed an acoustic sensor which computes third-octave spectrograms on the fly instead of recording audio waveforms. For sample-efficient machine learning, we have repurposed a pretrained audio neural network (PANN) via spectral transcoding and label space adaptation. A small-scale study in a neonatological intensive care unit (NICU) confirms that the time series of detected events align with another modality of measurement: i.e., electronic badges for parents and healthcare professionals. Hence, this paper demonstrates the feasibility of polyphonic machine listening in a hospital ward while guaranteeing privacy by design.

Machine listening in a neonatal intensive care unit

TL;DR

The feasibility of polyphonic machine listening in a hospital ward while guaranteeing privacy by design is demonstrated and a small-scale study in a neonatological intensive care unit confirms that the time series of detected events align with another modality of measurement.

Abstract

Oxygenators, alarm devices, and footsteps are some of the most common sound sources in a hospital. Detecting them has scientific value for environmental psychology but comes with challenges of its own: namely, privacy preservation and limited labeled data. In this paper, we address these two challenges via a combination of edge computing and cloud computing. For privacy preservation, we have designed an acoustic sensor which computes third-octave spectrograms on the fly instead of recording audio waveforms. For sample-efficient machine learning, we have repurposed a pretrained audio neural network (PANN) via spectral transcoding and label space adaptation. A small-scale study in a neonatological intensive care unit (NICU) confirms that the time series of detected events align with another modality of measurement: i.e., electronic badges for parents and healthcare professionals. Hence, this paper demonstrates the feasibility of polyphonic machine listening in a hospital ward while guaranteeing privacy by design.
Paper Structure (13 sections, 1 equation, 3 figures, 1 table)

This paper contains 13 sections, 1 equation, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Acoustic sensor
  4. Third-octave spectrogram
  5. Spectral transcoder
  6. AudioSet classification with PANN
  7. Label space adaptation
  8. Application
  9. Deployment in a neonatal intensive care unit
  10. Visualization of sound events
  11. Proof of feasibility for continuous monitoring
  12. Conclusion
  13. ACKNOWLEDGMENT

Figures (3)

  • Figure 1: Flowchart of stages in the proposed approach. The first two stages are performed "on the edge". The last three stages are performed "on the cloud", i.e., on a central server.
  • Figure 2: Spectrograms from audio recordings in the Neonatal Intensive Care Unit (NICU). First row corresponds to audio recordings transformed into fast third-octaves spectrograms. Second row corresponds to Mel spectrograms transcoded with the transcoder. Third row corresponds to groundtruth Mel spectrograms, obtained with Mel transformation on the waveform. PANN-1/3oct predictions, using the mapping between Audioset classes and NICU classes, are shown in fourth row.
  • Figure 3: Presence of conversation and footsteps on a day of April 2023 in one room, as averaged over three-minute intervals. The badge of the health professional (EPC) and of the mother are also shown during the period. The shaded areas denote intervals in which more than one adult is present in the room.