Thanks for the suggestions. Below a brief summary of the responses I received. These came in three flavors:
1) solutions suggesting specific hardware at the client side (e.g. using e.g., a two channel audio card)
2) solutions using client side software (e.g., _javascript_)
3) offline and/or post-processing solutions
For our purpose (relatively large-scale online rhythm production experiments) solution type 1 is unrealistic.
[input from Werner Hemmert and others]
Solution type 2 was tried by several researchers/institutes (using, e.g, PsychoPY _javascript_, etc.). However, most report - as expected - relatively large timing errors, largely due to keyboard scan rates, drivers, and/or operating system (as reported in the references mentioned in the original message). (Despite the claim of
psychopy.org of <4ms precision in online studies).
[Input from Ignacio Spiousas, Nick Haywood, Ben Schultz, Kyle Jasmin and others]
N.B. PeerJ recently published a comparative study [1]
Solution type 3 was suggested by some: i.e. o record the rhythmic pattern by tapping e.g. with a pencil on your desk or device microphone, along with the streamed sound, at the client side, upload the resulting audio file using a standard browser, and analyse it at the serverside using onset-detection and some crosscorrelation techniques. Depending on the sampling rate, latencies can be reduced to 1 ms or less.
[Input from Roger Dannenberg, Krzysztof Basiński, Justin London and others]
N.B.1 Ben Schultz announced to make his version of Solution type 1 available as open source (repeated below).
N.B.2. Nori Jacoby announced to make their version of Solution type 3 available as appendix to a forthcoming paper (repreated below).
Nevertheless, my hope is still on some elegant solution of type 2. If you have one, please let us know.
Best,
Henkjan Honing
University of Amsterdam
Faculty of Humanities
Faculty of Science
.