In previous experiments it has been shown that reverberation affects the accuracy of onset detection and instrument recognition. Pre-processing a reverberated speech signal with dereverberation for automatic speech recognition (ASR), where reverberation also decreases efficiency, has been proven effective for mitigating this performance decrease. In this paper we present the results of an experimental study addressing the problem of onset detection and instrument recognition from musical signals in reverberant condition by pre-processing the audio material with a dereverberation algorithm. The experiments include four different onset detection techniques based on energy, spectrum and phase. The instrument recognition algorithm is based on line spectral frequencies (LSF) and k-means clustering. Results show improvement in onset detection performance, particularly of the spectral-based techniques. In certain conditions we also observed improvement in instrument recognition.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.