A "query by humming" system that would let you find a song by humming part of the tune.
The idea is simple: hum to the mic, record your hum, extract certain ting'alings corresponding to the melody and rhythm characteristics, then compare these features to those of the songs in your previously unmentioned database, then return to a ranked list of the songs or song segments most similar to the humming.
There are various technques to look at when trying to approach the damned idea, but basically, apply time series methods to match the pitch contour of the humming to the MIDI files in your database.
The Goal
To create a reliable, efficent and competantly large-scale system that collects thousands of songs and responds in seconds.
The system's accuracy, efficiency, and scalability are all issues that have created stumbling blocks..
On the most basic level.. even using the voice recognition on my phone has been a complete flop, but with different time series methods, signal processing, pattern recognition, database indexing, linguistics, we're looking to make it happen.
DOT
No comments:
Post a Comment