A Comparative Latency Study of Hardware and Software Pitch-trackers

In this empirical study, results show that recently developed software pitch-trackers can be as fast as some current commercial hardware pitch-trackers. Since latency was the most crucial factor in the performability of the ZETA system, comparative analyses were made of the ZETA and the AXON interfaces, as well as the fiddle~ object, a real-time audio analysis tool in the Opcode’s MAX/MSP environment on a 400MHz G3. A RetroPak-fitted ZETA violin was used to provide the audio input. The audio signal emitted by the ZETA pickup was sent independently to the ZETA VC–225 Violin MIDI Controller interface, the AXON AX–100 guitar interface, and the fiddle~ object, to be analyzed by their pitch-tracking algorithms. Both plucked and bowed (legato) articulations of the violin’s open G and E strings were used as stimuli. Once the pitch was analyzed, the interfaces and the fiddle~ object triggered a MIDI xylophone sound sample on a K2000R synthesizer. Both the acoustic audio signal and the output of the synthesizer were simultaneously recorded, then compared for latency times to see how fast the pitch-tracking algorithms were able to perform the pitch-to-MIDI conversion. Results show that the pitch-tracking of the fiddle~ object on a G3 is as fast as the AXON. The ZETA, by comparison, is slower than the fiddle~ object and the AXON. For high-pitched sounds, the average latencies were 15–35ms for the AXON, 15–45ms for the fiddle~ object, and 15–45ms for the ZETA. For low-pitched sounds, the average latencies were 20–55ms for the AXON, 25–100ms for the fiddle~ object, and 30–70ms for the ZETA. In all cases, the pitches were correctly identified.