Implementation of a monophonic note tracking algorithm on Android

摘要

Pitch tracking algorithms have been proposed in many digital speech processing literature. Among the practical use of pitch tracking are: improved recognition, improved speech synthesis, and semantic disambiguation. A similar problem to pitch tracking when applied to music input signals, is note tracking, i.e. detecting all the notes in the perceived music. The general problem of music recognition seems to be beyond the techniques that have been accomplished by the advances in digital speech processing. A “real music” signal is composed of multiple sound from several instruments, and digitally separating the mix into individual channels/tracks is a hard problem to solve. The algorithm described in this paper assumes that the input signal is produced by a single source and further it focuses on monophonic sound, as opposed to polyphonic sound where two or more notes are played at the same time. The algorithm described below has been implemented on an Android device using proper building blocks (Activity and Service) that comply with the Android design guidelines to achieve the best performance. In addition to the standard Android libraries from the latest Android SDK, the application also relies on a third-party library for digital signal processing routines. The Android implementation of the algorithm has been tested using input sources from human voice and musical instruments. The paper also shows the experimental results of handling these input sources.