Understanding Audio

CCTV networks do not often have audio inputs. But when they do Autonomy Virage can incorporate advanced audio processing capabilities that can distinguish speech, gun shots, or sudden noises in the vicinity of the camera. The system can then automatically generate a report, alert, or change the camera position in response.

Autonomy Virage speech recognition technology also enables security organizations to search files in video, radio and telephony systems instantly. Autonomy’s speech technology is fundamentally different because it leverages IDOL’s conceptual understanding of content; whereas other technologies adopt a simple phonetic approach using only acoustic information, Autonomy achieves a higher level of understanding through language modeling.

Language modeling involves concept extraction in conjunction with acoustic-phonetic methods to achieve significantly greater accuracy and better results. Simple acoustic-phonetic methods alone fail to achieve good speech to text translation. The acoustic-phonetic approach cannot differentiate, for example, between “can I” and “can eye”. In this example, where the desired option is “can I”, Virage’s speech technologies employ IDOL’s intelligent probabilistic language modeling to understand the context of what is being said and in this way select the appropriate option “can I”. Virage’s audio recognition functionality includes:

• Speaker independence: the system is trained on a large balanced corpus of data encompassing many different variables such as different accents or male-female pitch and tone - this means the acoustic models are speaker independent. The solution works out-of-the-box with no manual training although customization for specific accents or speakers can be done;
• Extensive vocabularies: there is no arbitrary limit on vocabulary size - Speaker identification: audio recognition can be trained to enable individual speakers to be identified - Word spotting and phrase recognition: audio can be searched by standard keyword as well as conceptual methods. Conceptual searching returns references to conceptually related information ranked by relevance or contextual distance;
• Patented Autonomy technology: reduces CPU and memory usage for increased speed of operations and improved performance - Support for both high quality audio such as broadcast as well as telephony.