The Influence of Sampling Frequency on Guitar Chord Recognition using DST Based Segment Averaging

Generally, a sampling frequency of chord recognition systems follows the Shannon sampling theorem. This paper investigates the influence of sampling frequency that does not follow the Shannon sampling theorem in a chord recognition system. The chord recognition system uses a transform domain approach that makes use a DST (Discrete Sine Transform). In addition, the chord recognition system uses segment averaging for feature extraction. An acoustic-electric guitar is used in this work. Based on the experiments, a decrease in the sampling frequency from 2500 Hz down to 156 Hz has almost no influence in the recognition rate. As a first note, we only consider the recognition rate above 95%. As a second note, we use at least frame blocking length of 128 points and feature extraction length of 16 points. Therefore, if that kind of recognition rate could be accepted, the sampling frequency as low as 156 Hz could still be used for guitar chord recognition.