Abstract:
The semantic dimension of rough/smooth can be an important factor in customer preference of automotive powertrain sound. With this in mind, a new model is proposed for predicting roughness of powertrain sound. The model performs a signal decomposition into critical bandwidth channels, specific roughness prediction for each channel, and a combination of specific roughness across frequency to yield an overall roughness measure. Signal decomposition is performed with a bank of overlapped, critical bandwidth, bandpass filters. Because the signal in each channel is well represented by a small number of harmonically related, narrow-band components, it is assumed that specific roughness can be determined from the peaks in the magnitude spectrum of the envelope signal. A masked threshold estimate from the input signal is used to predict audibility of envelope fluctuation in each filterbank channel. Channels whose envelope fluctuation falls below threshold are assumed not to contribute to overall roughness, and are therefore discarded. Specific roughness values of the remaining channels are combined with a power-law model, yielding an overall roughness measure [E. Terhardt, Acustica 30, 201--213 (1974)]. Results of subjective tests of powertrain sounds are presented which validate the performance of the model. [Work supported by Ford Motor Company.]