(An)Adaptive Voice Query Transcription Scheme for Music Information Retrieval
http://www.riss.kr/link?id=T11281572

-
-
Accurate music transcription is essential in the voice-based (such as humming) music retrieval. Even though many researches have been done so far to develop pitch tracking and musical onset detection algorithms for accurate music transcription, the re...
-
Accurate music transcription is essential in the voice-based (such as humming) music retrieval. Even though many researches have been done so far to develop pitch tracking and musical onset detection algorithms for accurate music transcription, the results are not still satisfactory.
In this thesis, a new music transcription scheme is present with following new features: (i) for accurate note onset/offset detection, a new method WAE (Windowed Average Energy) was proposed. WAE defines multiple small but coherent windows with local threshold values; (ii) for accurate ADF onset detection, Dynamic ADF defines instead of traditional single threshold based methods using DTC (Dynamic Threshold Curve); (iii) for accurate acquisition of fundamental frequency of each frame, CAMDF (Circular Average Magnitude Difference Function) was used; and finally (iv) for the frequency realignment, K-means clustering algorithm was applied.
To evaluate the performance of the proposed scheme, prototype music transcription system called AMTranscriber (Automatic Music Transcriber) was implemented and various experiments were carried out. Experimental result shows that the proposed scheme can improve the transcription accuracy up to 95%. And also, mobile music information system named M-MUSICS (Mobile MUsic Semantic Indexing and Content-based retrieval System), which is based on GA (Genetic Algorithm) and RF (Relevance Feedback), was implemented. By this implementation, it is proved that the scheme is applicable in music information retrieval system.
-
초록전체보기