A three year project funded by the Engineering and Science Research Council (EPSRC), UK. under grant number GR/S77530/01. Completed 8/04 - 7/07.
This project was concerned with places for music performance or reproduction, places such as auditoria for classical music and arenas for pop music. Of secondary interest, are spaces where speech intelligibility is important. Trained individuals can listen to music played in a room and estimate the acoustic quality of the space. This project has been exploring whether it is possible to get a computer algorithm to mimic this judgement? This is not just of academic interest, if room acoustic parameters could be derived from reverberated music, it would enable measurements to be easily made using the everyday signals in a space, forming a non-invasive test method for occupied, in-use room measurement.
A great deal of room acoustics’ research involves the measurement and characterization of acoustics in terms of parameters such as reverberation time. A large body of knowledge has been built up, enabling designers to work towards target parameter values. Unfortunately, the measurements that the design criteria are based upon are limited to unoccupied cases and it is known that the acoustic is altered by occupancy. Occupied measurements usually have poor accuracy and are generally unsuccessful, consequently, reliable occupied data are not generally available. It would be an important advance if a system could be devised to enable occupied, in-use measurements to be routinely made. This could lead to a better understanding of acoustic requirements, improved design criteria and consequently better buildings. The need for occupied measurements isn’t new, but previous methodologies are insufficiently accurate or robust. The timeliness of the project comes from trying to solve an old problem by applying modern tools.
In addition to room measurements, blind estimation techniques are potentially useful in some state-of-the-art signal processing systems. The idea is that these systems will use the estimated acoustic parameters to adapt themselves to the environment, for example, an estimated reverberation time might be used by speech cleaning pre-processors before automatic speech recognition, blind channel equalisation in communication systems or intelligent hearing aids. Due to these new applications, estimation of reverberation time from naturally occurring sounds has attracted increasing attention during this grant.
The system proposed derives the parameters from a music signal received by a microphone in a room. One system uses stochastic machine learning to extract determining characteristics from the reverberated music signals and thereby estimate the parameters. A second method uses a Maximum Likelihood Estimation (MLE) method applied to the gaps between music notes where free decay is evident. In this way, the systems measure parameters such as reverberation time, IACC and clarity.
Investigators: Trevor Cox and Francis Li (Salford) and Jonathon Chambers (Cardiff)
Researchers: Paul Kendrick (Salford) and Yonggang Zhang (Cardiff)
Quantifying Room Acoustic Quality Using Artificial Neural Networks
A three year project funded by the Engineering and Science Research Council (EPSRC), UK. under grant number GR/L34396. Completed May 2001. Rated “tending to outstanding”
Principle Investigator: Prof. Trevor Cox; Research Assistant: Dr Francis Li