A novel algorithm based on time-frequency analysis for extracting melody from human whistling

Human whistling signal can be considered as a potential human-computer communication media. Besides, this signal can be used in music production, where music composers whistle their melody and an accurate note extraction system converts their whistled melody into musical notes. Thus, this paper introduces an accurate and robust note extraction system called SFMS. This system is based on four stages. First, in the Segmentation stage, starting and ending indexes of each note will be extracted. In the next stage, named Frequency extraction, dominant frequencies of notes will be calculated and in the Mapping stage each frequency will be transformed into a note number. Finally, in the MIDI Synthesis stage the output will be synthesized into a MIDI file. The SMQT transform is utilized in the Segmentation stage to add robustness to this process. Since the first stage is a robust segmentation algorithm, frequencies are extracted with a full time frequency analysis which results in an accurate dominant frequency estimation and extraction. Experiments are conducted and discussed. We have introduced of method to measure the musical error of note extraction algorithms and the results show that our algorithm works with 92.7% accuracy.