Time-scale modification of music using a subband approach based on the bark scale

Abstract

Time-domain time-scaling algorithms are efficient in comparison to their frequency-domain counterparts, but they rely upon the existence of a quasi-periodic signal to produce a high quality output. This requirement makes them unsuitable for use on multi-pitched signals such as polyphonic music. However, time-domain techniques applied on a subband basis can resolve the multi-pitch problem. We propose an improved subband implementation based upon the bark scale for the timescale modification of music. The new subband approach is supported by psychoacoustic and music theory and subjectively through informal listening tests.

Time-scale modification of music using a synchronized subband/time-domain approach

“…The test used time-scale factors ranging from 0.66 to 2 and all tracks were sampled at 44.1kHz. The 'non-synchronized' tracks were time-scaled using the same parameters given in [3]. The 'synchronized' tracks were partitioned into subbands using the same cutoff frequencies as in [3], with SR and L stat set to 20ms and 33ms, respectively, for all subbands.…”

“…The 'non-synchronized' tracks were time-scaled using the same parameters given in [3]. The 'synchronized' tracks were partitioned into subbands using the same cutoff frequencies as in [3], with SR and L stat set to 20ms and 33ms, respectively, for all subbands. All thresholds were set assuming the maximum amplitude corresponded to 90dB(SPL).…”

“…The solutions to these issues are diametrically opposite since partitioning a complex waveform into many subbands reduces the complexity of each subband but increases potential subband synchronization problems and vice versa. While [1] and [2] partition the complex signal into subbands using uniform width filterbanks, [3] justified the use of a non-uniform width filterbank based upon the bark scale for the improved time-scale modification of Western tonal music. However, subband synchronization still remains an issue.…”

“…One problem with the subband/time-domain approach is that a reverberant/phasy artifact is introduced into the time-scaled output due to poor synchronization between time-scaled subbands, as explained in [3]. Lack of synchronization between subbands is also noticeable during 'hard' transients, resulting in the time-scaled transients sounding metallic and harsh.…”

“…The discard/repeat process relies heavily upon the existence of a quasi-periodic waveform, making timedomain approaches suitable for speech and monophonic music but unsuitable for their direct application to most polyphonic music due to the generally complex multi-pitch nature of these types of waveforms. However, [1], [2] and [3] have demonstrated that applying time-domain time-scale modification algorithms on a subband basis can resolve this issue.…”