[Olympus developers 172]: Re: Audioserver VAD
Antoine Raux
antoine.raux at gmail.com
Wed Nov 18 14:17:36 EST 2009
Hi Jose,
The settings for AudioServer are in the AudioServer.cfg file, which is
located in the Configurations folder you're using (i.e. under
<YourSystem>/Configurations/<YouConfigName>).
Below is the example for the LetsGoPublic system.
You can see there are several possible VADs (energy-based vs GMM based)
so it depends which one you're using. The energy-based VAD can be tuned
by changing the power_threshold value.
If you set "log = 1" (as is the case here), you should see the kind of
values you're getting for energy (at least I think so). Based on those,
you should be able to set the threshold. Note that another option is to
tune your audio device input volume. For GMM-based VAD, I don't think
there's an easy way to tune it (short of retraining the GMMs on your own
data).
hope this helps.
antoine
# This is a configuration file for the Audio_Server for LetsGoPublic
# Sample rate
sps = 8000
# Host and port number of Sphinx recognition engines
engine_list = male:localhost:9990,female:localhost:9991,dtmf:localhost:9992
# Enable/disable logging
log_full_session_input = 1
log = 1
# Configuration for the Voice Activity Detector
#vad = power
#vad_config = window_width=800,power_threshold=10000
vad = GMM
vad_config =
model_dir=../../Resources/GMMVAD/letsgo.5class.endptr.ci_continuous_1gau,
sampling_rate=8000, fe_frame_rate=100, fe_window_length=0.0256,
fe_fft_size=512, fe_num_filters=35, fe_lower_filter_freq=130,
fe_upper_filter_freq=3800, fe_normalize_c0=1, prior_noise_level=8,
prior_speech_level=15, snr_estimation_step=80000, window_width=20
José David Lopes wrote:
> I am trying to connect another ASR to Olympus. I found some
> difficulties because VAD that comes with Audioserver is very
> sensitive. It triggers both speech and pause events very quickly.
> Is it possible to tune the sensitivity of this parameter?
>
> José David Lopes
>
More information about the Olympus-developers
mailing list