Hey @sahil.dutta for the available parameters, in terms of the noise cancellation models like Quail Voice Focus you could play around with the enhancement level, which is basically a mix of the raw and enhanced audio - STT models are generally trained on the real noisy audio, so some degree is fine - if it’s over suppressed it may introduce some artifacts. I would suggest playing with the values around 0.5-0.85, but it really depends on your case, as well as the STT or real-time models that you are using. Similar for the VAD, where you have 3 parameters:
* sensitivity: where VAD 2.0 is probability based other models would work on the energy level. Nevertheless, it mostly comes down to how loud the speaker must be to trigger VAD, you can start from 0.5 and go up to 1.0 where 1.0 would almost never trigger VAD as it would need 100% probability of the speech.
-
minimum_speech_duration -how long does the VAD still keeps triggering after the speech stop(50-100ms could be good starting point) -
minimum_speech_duration -like the same says for how long you need to keep the speech to trigger VAD (200-300ms could be good starting point)For the pricing, if you are self-hosting, most likely you are using our SDK from the dev platform. Feel free to test it for the first 30 days, and we will reach you out
Also more than happy to connect you with the team.
To sum up, it really depends on your setup like what STT engine are you using, what is the acoustic setting, distance from the microphone etc. I can highly recommend visiting our developer playground: https://developers.ai-coustics.com/dashboard/sdk/playground where you can either bring your own audio recording, or test it live with various settings.