Low Latency Audio Pitch Shifting in the Time Domain

This page is a complement to the scientific paper "Low Latency Audio Pitch Shifting in the Time Domain", published in the proceedings of the IEEE International Conference on Audio, Language and Image Processing (ICALIP 2008). It provides music excerpts to illustrate the proposed algorithm (the "Rollers" algorithm) and to compare it with other well-known approaches. A pitch shifting of +30% is used for all algorithms.

Original music Time-domain overlap-add WSOLA Standard phase vocoder Rollers Improved phase vocoder
Latency (theoretical): 30ms 100ms 46ms < 10ms 93ms
Listen Listen Listen Listen Listen Listen
Listen Listen Listen Listen Listen Listen
Listen Listen Listen Listen Listen Listen

Sources

The excerpts are taken from the following musics:
Music 1: The Coming Through, by Matthiew Lien, in OpenMusic CD.
Music 2: Trend, by Jason Honingford, in Aminet
Music 3: The March of the Globins, by Tobias Steinmann, in OpenMusic CD

The "Rollers" algorithm uses several frequency shifting transformations to express a pitch shifting transformation. Frequency shifting and pitch shifting are different audio effects. The following music excerpts illustrate the difference between frequency shifting and pitch shifting.

Note that we use the term "pitch" to refer to the musical pitch (measured in octaves and cents rather than in Hz). With this definition, we have pitch ~ log2(frequency). However the pitch shifting ratio is given by the amount by which the frequency is multiplied.