Barry, I'm glad to hear Bpod is becoming "standard" at Cold Spring Harbor!
Currently, the PsychToolboxSoundServer plugin for Bpod comes close to an ideal solution in several ways:
It can generate stereo 192kHz sampled sound, with ~7-8ms latency, using the Asus Xonar DX sound card.
Loading sounds onto the server is very fast compared with solutions that move raw sound data over USB or Ethernet.
The 7-channel sound card provides a 1ms TTL signal concurrent with sound onset, on channels 3-5. Using this part and a cable included with the sound card, the TTL can be routed back to Bpod, for a ~100?s precise measurement of actual sound onset on each trial.
The specs we released on the wiki were measured on Ubuntu 14 configured with a low-latency core. We recently discovered that Windows 7 computers can also have latency as low as 7-8ms with ~1ms jitter, when using the latest ASIO driver from Asus and setting the buffer to 1ms in the ASUS control panel.
Areas that need to be improved to make the sound server truly ideal:
The sound card resides on the computer that runs Bpod. This means that processor-hungry background tasks can cause extra latency and jitter in some configurations - so a multi-core processor (read: intel core-i7) is pretty much mandatory. Even with multi-core, you have to be careful about what apps you have running apart from MATLAB. In the worst situations I have seen, only a very small fraction of trials (1-2%) are affected with extra latency of ~10-20ms, and the TTL feedback allows you to account for the exact onset time in your data. Nonetheless, it is not ideal because it limits the possible configurations of the computer.
A stand-alone sound server running on a dedicated computer with a real-time operating system (or a microcontroller) would solve this problem. We have software for a lower-resolution (44.1kHz Stereo) microcontroller-based sound server available here, but nothing that can play ultrasound yet. At Sanworks, we have a HiFiBerry in the workshop, courteously donated by Zador Lab. Unfortunately, Raspberry Pi 3 still does not have gigabit Ethernet - so we would have to find a fast networking solution to avoid long inter-trial sound transfer delays.
We would be thrilled to work together with someone in your lab (or elsewhere), who would like to solve the problem of low-cost, low-latency, high-definition sound for the community!
Concerning speakers, we are testing some low-cost human-audible speakers for our product line, but good ultrasound-capable speakers are hard to come by. I know of these from TDT, and these from Avisoft. Anyone who has input here would be more than appreciated!