SPIRIT VAD (Voice Activity Detection) detects the presence of speech in the
signal. It has special adaptive algorithm to automatically adjust to the level of
the noise in the signal, in order to provide robust operation even in the noisy
environment. It has many user configurable parameters, allowing the algorithm
to optimally tune itself for a specific application. VAD outputs several
coefficients (up to 10) that characterize spectral envelope of the noise (when
no speech is detected), so that the regenerated noise would be similar to the
original noise.
SPIRIT CNG (Comfort Noise Generator) generates noise, distributed either
uniformly or shaped according to the spectral envelope coefficients, which are
passed to CNG as parameters by VAD.
Features
Direct interface with PCM 8KHz sampled data. Both sample-by-sample and block based processing are supported
Adaptation to background ambient or line noise
Voice activity classification as "early" or "sustained" according to specified parameters
Adjustable adaptation time, hangover period and thresholds
VAD generates LPC coefficients describing a noise spectrum envelope. CNG accepts those coefficients and recreates the noise with a similar spectrum
eXpressDSP compliant. Code is reentrant, supports multithreading and dynamic memory allocation. At the same time allows direct (noneXpressDSP) interface to enable static memory allocation
Benefits
Highly optimized code ideal for resource constrained applications
