Query by Humming, or QbH, can be defined as a system that allows users to find a song simply by humming its melody. This melody acts as an input query, and once it enters the system, it is compared to a large database of other melodies. Notes and rhythms are the basic characteristics used by the system to recognize the song. Hummed by the user, the melody is recorded, and these characteristics are extracted to be compared to the features of other songs. The system operates in a straightforward way: after the queries are hummed into a microphone and converted into a digital format, they enter a pitch-tracking module (Mackenzie, 2000). The system then creates a contour image of the melody, and this representation is transferred in the query engine, which is built to use an approximate pattern algorithm to tolerate certain inaccuracies. Finally, a list of the melodies with matching patterns is produced and presented to the user (Dickson, 2003). This research relates to the Music Signal Processing field, and music information retrieval in particular, and its aim is to identify the way to create an efficient, reliable large-scale database of songs.
While a significant number of research projects have been done on the topic of music retrieval systems, many problems still exist, as the field is immature. For example, the systems are not robust to poor humming, noise, and distortion (Li et al., 2010). The majority of these projects only involve simple, monophonic music, and there is a substantial lack of methods of evaluation. Moreover, all of the projects have limited database capabilities that may be too specific and do not address the need for large-scale databases (Davenport, 1993). The majority of other issues within the query by humming research, such as wrong pitch, note duration, or key, are related to the nature of music itself and the way people perceive and recognize music. These issues must also be addressed in future research to create a reliable system containing tens of thousands of songs and responding quickly to queries.
As the field grows, the literature on the topic of Query by Humming research is increasing rapidly as well. One example of such research is a project by Parth Patel. In his project, Patel discussed different strategies to retrieve music and analyzed the Query by Humming approach in detail (2019). He also reviews various information retrieval architectures, speech feature extraction, machine learning-based music retrieval, and audio fingerprinting (Patel, 2019). Patel proposes a dataset for the system and describes the technical stack, data processing, database preparation, and storage involved in the development of such systems. The experiments conducted during the course of development are presented in the paper and discussed in detail as well. Another project by Bradshaw uses qualitative and quantitative analysis to study music information retrieval systems and how or if they can be used in music education classrooms (2017). Discussing the background of the topic, he also reviews several examples of research literature (Bradshaw, 2017). Bradshaw has conducted his research through three rounds of surveys of a group of music educators, completing a descriptive analysis afterward.
There is research that addressed the issues and problems of the Query by Humming technology as well. Khan and Mushtaq have highlighted these challenges and discussed how they could be solved to ensure further advancement of the field (2011). They supported this discussion by conducting surveys as well and described four main categories that Query by Humming field can be divided into, depending on the technique used. The first technique is String Matching, which occurs when a melody is extracted from the song, converted to the audio signal, and then assembled to its contour image. The second technique is called the Tree-Based Search Technique. It stores contour representation in a tree structure, applying brute force mechanisms to gather the results and provide the information on the query. The third technique is Dynamic Time Warping Technique, a standard algorithm inherent to audio input data (Kim & Jang, 2002). The fourth technique is a Hidden Markov Model that functions by looking into all possible options for comparison. This technique has been adapted for Query by Humming as a statistical approach with two modules designed to segment the notes into signs and signals.
The music industry has changed drastically in the past several decades and is continuing to develop and grow. Human perception and cognition of music are interconnected with these developments, and the way people listen to and think about music significantly depends on it. The need for advanced technologies that would make Query by Humming systems available and efficient is only expected to grow (Wijnberg et al., 2002). This is mainly because people want to be able to easily find music without knowing or remembering its title or any other characteristics, and this technology will allow them to do it in seconds. Although the 100 percent accuracy in tracing humming has not been achieved yet, further research is needed to consider and create ideal circumstances to improve the accuracy of the searched results.
References
Bradshaw, B. (2017). How Query by Humming, a Music Information Retrieval System, is being used in the music education classroom. Journal of Multimedia Information System, 4(3), 99-106.
Davenport, T. H. (1993). Process innovation: Reengineering work through information technology. Harvard Business Press.
Dickson, P. R. (2003). The pigeon breeders’ cup: A selection on selection theory of economic evolution. Journal of Evolutionary Economics, 13(3), 259-280.
Khan, N. A., & Mushtaq, M. (2011). Open issues on query by humming.Fourth International Conference on the Applications of Digital Information and Web Technologies, 147-152.
Kim, S., & Jang, K. (2002). Designing performance analysis and IDEF0 for enterprise modelling in BPR. International Journal of Production Economics, 76(2), 121-133.
Li, J., Han, J., Shi, Z., & Li, J. (2010). An efficient approach to humming transcription for query-by-humming system.2010 3rd International Congress on Image and Signal Processing, 45-50.
Mackenzie, K. D. (2000). Processes and their frameworks.Management Science, 46(1), 110-125.
Patel, P. (2019). Music retrieval system using query-by-Humming. Master’s Projects, 895.
Wijnberg, N. M., Van den Ende, J., & De Wit, O. (2002). Decision-making at different levels of the organization and the impact of new information technology.Group & Organization Management, 27(3), 408-429.