It often happens that one remembers the tune of a song but either fails to recall or perhaps is unknown of the song's details such as title, lyrics or the artist's name. In such situations, searching the song in a search engine becomes quite difficult by using the textual query approach. At such a time it is most desired that if one could just hum the song's tune and could get the desired song in the results.
This is possible as there exists systems that retrieve results from an audio database based on humming.
QUERY BY HUMMING SYSTEM
Query by humming(QbH) is a music retrieval system that breaks the traditional way of classifying songs by title, genre , artist or composer.
Natural way of querying a textual database is to provide a text query. Similarly the most natural way to query an audio database is with a humming input.
User provides query by humming it in the microphone.The query is then digitised and passed on to a pitch tracker module. Pitch tracker converts humming into a sequence of relative pitch transitions and outputs a contour representation of the input melody. Different methods for tracking pitch are :
Figure below is a representation of the discussed system -
This is possible as there exists systems that retrieve results from an audio database based on humming.
QUERY BY HUMMING SYSTEM
Query by humming(QbH) is a music retrieval system that breaks the traditional way of classifying songs by title, genre , artist or composer.
Natural way of querying a textual database is to provide a text query. Similarly the most natural way to query an audio database is with a humming input.
User provides query by humming it in the microphone.The query is then digitised and passed on to a pitch tracker module. Pitch tracker converts humming into a sequence of relative pitch transitions and outputs a contour representation of the input melody. Different methods for tracking pitch are :
- Autocorrelation
- Maximum likelihood
- Cepstrum Analysis
Figure below is a representation of the discussed system -
The approach is based on the observation that a sequence of relative differences between pitch of consecutive notes can be used to distinguish between melodies. The method uses an alphabet of three possible relationships between pitches - 'U' representing situation where a note is above the previous note , 'D' representing situation where a note is below the previous note and 'S' representing situation where a note is same as the previous note.
TUBENOT
Tubenot is a query by humming search engine developed by Arefin Huq, Mark Cartwright and Bryan Pardo of the Interactive Audio Lab at Northwestern University.
Users can search the database by humming into microphone or by playing melody on the virtual keyboard or by typing some of the lyrics to find the desired song.
It compares the hummed query to the songs in database by considering the note intervals. Moreover, it compares the query with potential targets by considering the rhythmic ratios between notes.
Tubenot learns from user input and improves the results as more queries are generated by the user. Tubenot asks the user to mark the desired song from the results and then pairs the song with the user's query and runs a genetic algorithm to control how the system compares user's query with target.
Other examples of Query by Humming System are -
CHALLENGES INVOLVED
- ACRCloud SDKs
- SoundHound (mobile app)
- MaART at Sourcsforge
- Query by humming system from Musipedia
CHALLENGES INVOLVED
- Signal processing required for extracting melody from the stored songs and query.
- Application of appropriate pattern matching algorithms for returning proper ranked results.
- System must be able to handle inaccuracies in user's singing.
- System must operate on reasonable size database and must be able to retrieve results fast.
APPLICATIONS
Query by humming system can be used by an individual for his/her personal use. An individual may use it to search for a song that he/she may have heard somewhere such as in a passing by car on the street. The user can find the song without any knowledge of the song's title or singer.
QbH systems also have applications in commercial music industry, music radio and TV stations and music stores.
Query by humming system can be used by an individual for his/her personal use. An individual may use it to search for a song that he/she may have heard somewhere such as in a passing by car on the street. The user can find the song without any knowledge of the song's title or singer.
QbH systems also have applications in commercial music industry, music radio and TV stations and music stores.
References
[1] Query by humming Musical Information Retrieval in an Audio Database
- Asif Ghias ,Jonathan Logan David Chamberlin Brian C. Smith (Cornell University) http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.42.7644&rep=rep1&type=pdf
[2] https://en.wikipedia.org/wiki/Query_by_humming
[3] https://en.wikipedia.org/wiki/Tunebot
[4]https://www.ee.iitb.ac.in/student/~daplab/publications/internationalconference/papers/ar-pr-ncc03.pdf
[1] Query by humming Musical Information Retrieval in an Audio Database
- Asif Ghias ,Jonathan Logan David Chamberlin Brian C. Smith (Cornell University) http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.42.7644&rep=rep1&type=pdf
[2] https://en.wikipedia.org/wiki/Query_by_humming
[3] https://en.wikipedia.org/wiki/Tunebot
[4]https://www.ee.iitb.ac.in/student/~daplab/publications/internationalconference/papers/ar-pr-ncc03.pdf
Comments
Post a Comment