Blind Estimation of Room Volume from Reverberant Speech Based on the Modulation Transfer Function

Estimating room volume is crucial in acoustic engineering, architectural acoustics, and emergency response. Traditional methods relying on the Sabine equation face challenges due to the difficulty in accurately measuring absorption coefficients. This paper introduces a method for blindly estimating room volume using reverberant speech on the basis of the modulation transfer function (MTF). Our approach processes reverberant speech through a seven-octave band filterbank, extracting temporal amplitude envelopes to form MTF features and input the MTF feature into a convolutional neural network to estimate room volume. Experimental results on various real-world and simulated room impulse responses demonstrate our method’s effectiveness, achieving a mean square error of 0.14 in volume estimation. The mean absolute logarithm of the ratio between the estimated volume and the ground truth is 1.15. This approach significantly improves accuracy and robustness over the baseline method, offering a practical solution for scenarios where direct measurements are impractical. Potential applications include security monitoring, space management, and emergency evacuation planning.