The electret capsules used in the binaural head are Primo capsule EM273. They are rated for 80 dB signal to noise ratio or 14 dBA. These are good quality capsules and are very popular. The rating for 14 dBA looks like it is true because I was able to measure down to 17.8  dBA noise level for 231 ms duration part and 13.87 dBA for 1.93 ms part and it was not accomplished in the lab. It is taken from the demo record called "Leaves are falling". The record was taken in unusually quiet night with few sounds and between them I found the quietest part of the track which was 231ms in length.  From this part I took another quiet part of 1.93 ms.  For 231 ms part the measurement is accurate because it is quite long, but it has some environment noise, not only self noise. For 1.93 ms length the RMS calculation method is accurate down to 258 Hz (half wavelength). The noise spectrum of the capsule should be dominating somewhere around 10 kHz according to bksv hanbook. So 13.87 dBA can be close to reality. Of course for a short signal length like 1.93 ms probability comes into play for a random noise which is kind of rocket science. I think 13.87dBA probably is not a true number but somewhere close to it. 

   In addition ear canal increases the signal to noise ratio for another 3 dB as mentioned here. But I think it would be nice to have even better specs in the future. I see the possibilities for another 6 dB increase or so. But that would increase complexity and price of the binaural head.