Acoustical Quality Assessment of the Classroom Environment

Marian George, Moustafa YoussefarXiv 2012 · OpenReview

TL;DR

This paper introduces a novel system that uses machine learning to automatically assess classroom acoustics by analyzing sound features of both instructors and students. This approach offers a continuous, objective feedback mechanism for improving teaching strategies and learning environments, moving beyond traditional subjective surveys.

What Problem Does This Paper Address?

While the importance of effective teaching and optimal learning environments is widely recognized, objective and continuous assessment of classroom acoustics has been lacking. Existing research focuses on acoustical design recommendations or facilitating lecture review, but not on using real-time acoustical data to provide feedback on the learning process itself. Poor acoustics can negatively impact speech intelligibility, student learning efficiency, and even lead to voice disorders in instructors. This paper aims to bridge this gap by developing a system that can continuously monitor and evaluate the acoustical quality of a classroom, offering actionable insights for educators and institutions.

Methods

The core of the proposed system lies in its use of speech processing and machine learning techniques.

  • Feature Extraction: The system extracts several key audio features:
  • A-weighted Sound Pressure Level (Lp): This measures the overall sound intensity, crucial for distinguishing between noisy and quiet environments.
  • Pitch and RASTA-PLP Coefficients: These features are used for gender classification of speakers. Pitch, a measure of vocal frequency, is particularly useful, and RASTA-PLP (Relative Spectral Perceptual Linear Prediction) enhances robustness in noisy conditions.
  • Machine Learning Classifiers:
    • K-Nearest Neighbors (K-NN): Used to classify individual sound recordings as either "noisy" or "quiet." The lecture as a whole is then classified based on the majority of these recordings.
  • Gaussian Mixture Model (GMM): Employed for gender identification of speakers, leveraging the extracted pitch and RASTA-PLP features.
  • Noise Location Estimation: By placing multiple microphones at different locations within the classroom and analyzing the sound pressure levels, the system can approximate the source of student noise (e.g., front left, back right).
  • Correlation Analysis: The system investigates relationships between various factors, such as student noise levels, instructor speech levels, and instructor gender, using chi-square tests of independence and proportion comparisons.

Datasets

The experiments were conducted in four university classrooms. Recordings were made using Nokia N97 mobile phones. The data consists of short-interval (10-20 seconds) recordings taken throughout thirty live, one-hour lectures. For noise location estimation, controlled experiments involved placing four microphones and having students engage in side-talk for approximately 5 minutes. No specific benchmark datasets were introduced; rather, real-world classroom environments were utilized.

Key Insights and Ideas

  • Objective Assessment: Moving beyond subjective surveys, this system provides an objective and continuous measure of acoustical quality, enabling more consistent and timely feedback.
  • Instructor and Student Focus: The system uniquely analyzes features from both the instructor and students, recognizing that classroom acoustics are a dynamic interplay between them.
  • Actionable Insights: The ability to pinpoint noise locations allows instructors to proactively address disruptive behavior, while correlations between speech levels and noise provide guidance on optimal speaking volumes.
  • Teacher Well-being: By identifying noisy environments, the system can indirectly help reduce the vocal strain on instructors, potentially mitigating voice disorders.
  • Efficiency: Using automated classification speeds up the assessment process, allowing for continuous monitoring rather than infrequent surveys.

Key Results

  • Noisy/Quiet Classification: The k-NN classifier achieved a classification accuracy of 94.2% for distinguishing noisy from quiet lectures when using k=5.
  • Noise Location Estimation: The system demonstrated 100% accuracy in determining the approximate location of student noise (front left, front right, back left, or back right).
  • Instructor Speech Level and Noise: A high level of association (difference in proportions of 0.75) was found between lecture noise level and instructor's speech level. This suggests that instructors often raise their voice in response to background noise.
  • Instructor Gender and Noise: A weak association (difference in proportions of 0.4) was observed between lecture noise level and instructor's gender, aligning with previous studies suggesting gender bias is not a significant factor in student ratings.

Limitations and Future Work

  • Limitations:
    • The study acknowledges that the correlation between instructor gender and noise level is weak.
  • The system's current analysis of noise location is an approximation.
  • The paper notes that typical classroom noises (coughing, chair movement) were not removed, which might influence results.
  • Future Work:
    • The authors plan to evaluate the system in four university classrooms in the upcoming semester.
  • They intend to expand the extracted features to include other non-acoustical instructor characteristics.
  • Further research will focus on generating additional correlation results to enhance the learning environment.
  • The system aims to guide instructors in taking measures to improve learning environments, such as paying more attention to noisy corners or adjusting speech levels.

Summary updated

Abstract

Teaching is one of the most important factors affecting any education system. Many research efforts have been conducted to facilitate the presentation modes used by instructors in classrooms as well as provide means for students to review lectures through web browsers. Other studies have been made to provide acoustical design recommendations for classrooms like room size and reverberation times. However, using acoustical features of classrooms as a way to provide education systems with feedback about the learning process was not thoroughly investigated in any of these studies. We propose a system that extracts different sound features of students and instructors, and then uses machine learning techniques to evaluate the acoustical quality of any learning environment. We infer conclusions about the students' satisfaction with the quality of lectures. Using classifiers instead of surveys and other subjective ways of measures can facilitate and speed such experiments which enables us to perform them continuously. We believe our system enables education systems to continuously review and improve their teaching strategies and acoustical quality of classrooms.
BibTeX
@misc{george2012acoustical,
  author = {Marian George and Moustafa Youssef},
  title = {{Acoustical Quality Assessment of the Classroom Environment}},
  year = {2012},
  eprint = {1201.2902},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1201.2902},
}

Loading related content...

Figure