Voice Activity Detection (VAD)

Robust voice activity detection (VAD) is the precondition for effective and robust applications for audio monitoring, voice communication and transmission, voice control of technical devices as well as the realisation of additional algorithms such as active noise reduction, acoustic echo compensation, microphone arrays and many more. The components for voice activity detection (VAD) as well as active noise reduction (ANR) were developed to meet the demands for both the highest possible robustness against adverse ambient conditions as well as minimal memory or respectively processing requirements, and incorporate proprietary processes both in time and frequency range, some of which are patented. The algorithms adapt themselves automatically to new ambient conditions and characterise themselves through their high robustness. Disposability is very high, as the algorithms have been frequently tested in practical integrated applications for machine control, as well as being subject to continuous maintenance and quality control. 

For further information go to:

• Features [...more]
• Supportable platforms [..more}
• Additional information [...more]

Features

upwards

• Fast adaption to changing ambient conditions
• Robust detection of speech with low energy, e.g. fricative
• Well suited for noisy environments in combination with our noise reduction algorithms
• Particularly suitable for DSP implementation

Technical features:

• Robust, patented processes for voice activity detection (VAD)
• Reduces the system load during voice transmission and recognition
• Enabling technology for a broad range of speech-based applications
• Autonomous adaption to background noise
• Integration with noise reduction
• Minimal resource requirements (approx. 7 MIPS, 4 KB memory)
• Continuous adaption and real-time convergence (<200 ms)
• Reliable performance even with large background noise levels (SNR >0 dB)
• ANSI C with hardware abstraction layer, delivery as:

• Object code for specific DSP environments
• DLL or static library for WIN and LINUX environments

• Parameter interface for integration of VAD in the target application

Supportable platforms

upwards

• Blackfin ADSP 2153x
• ADSP 2191
• WINDOWS (DLL, static library)
• Linux (DLL, static library)

Additional information

upwards

In addition to the integrated adaption processes to background noise, the algorithms can be further customised to the necessary performance requirements via selecting the appropriate parameters or respectively automatic context recognition (e.g. switching between frequency and time voice activity detection, threshold adaptation, etc) You can profit from our many years of experience with algorithms in the areas of automotive and vehicle communication, telecommunication, building technology and consumer electronics.