NPES vocoder

The NPES analysis/synthesis algorithm is based on "natural" model of human vocal tract. In that model it is assumed that locally constant parameters for speech signal are fundamental frequency, voicing frequency, frequencies and energies of the formants. This parametrization lets to use NPES vocoder as common algorithm for speech preprocessing for tasks such as compression, speaker identification or speech recognition. more details

NPES vocoder characteristics

Signal class - speech of single speaker.
Sample rate of the speech signal - up to 16 KHz.
The band of formant frequencies - 80..3800 Hz.
The band of fundamental frequency - 50..Fs/2 Hz.
Speech transfer bitrate - fixed in 800..2400 bps diapason.
The quality of speech - 2.6..3.6 MOS.
Processing delay - 25 msec.
Build-in adaptive noise filtering.
Transformation of the speech parameters.
Robustness at errors in transfer channel.
Algorithm implementation - Written in ANSI C++. Our current NPES SDK is for floating point processors.

Application field of the NPES vocoder

Speech compression for transfer and storage.
Speech transformation. With our VoiceVary program it is possible to shift the voice pitch and to change the vocal tract size in "real time".
Text to speech.
Speech recognition.
Speaker identification.
Measure of pitch frequency. With our Fork program it is possible to tune up musical instruments.

ADSS filter

Our adaptive digital filtering method which is based on the difference of the spectral characteristics of the useful signal and the background noise. Parasitic noise can be harmonic or stochastic of arbitrary coloring. Our ADSS filter can filter the useful signal and noise out of this mixture. more details

ADSS filter characteristics

Signal class - speech, music.
Sample rate of the speech signal - up to 44 KHz.
White noise suppression - up to 10 dB.
Single tone suppression - up to 20 dB.
Processing delay - 20 msec.
Algorithm implementation - Written in ANSI C,C++. Our ADSS SDK is for floating- and fixed point processors. It is ready to port to any hardware platform.

Application field of the ADSS filter

Obtaining clear sound recordings. With our SoundClear program it is possible to obtain a very clear sound on a standard PC without having to use studio equipment.
Sound transfer systems (telephony, radio, TV, telecommunication).
Preliminary processing for tasks of compression, identification and speech recognition.

All represented technologies are original products of our group. They are better than existing analogs in many characteristics, have potential for improvement and are progressing now by us.