Government agencies and many other organisations have to handle huge amounts of recorded conversation for security reasons. According to some estimates, more than 50 per cent of voice traffic is recorded for this purpose - so big organisations are stacking up millions of hours of voice data a year.
Collecting this information is one thing: making use of it intelligently to combat terrorism, drugs trafficking and other crimes, is much more difficult.
Manual transcription, translation and analysis are slow and prohibitively expensive. And sometimes the results are too late. So how can technology help? Traditional speech-to-text search, also known as Large Vocabulary Continuous Speech Recognition (LVCSR), systems are a big advance over listening manually. But they too have their limitations.
By their nature, LVCSR systems are computer processing power intensive and lack scalability. On grounds of cost and timeliness, speech-to-text systems are able to focus on only a small sample of recordings, typically around 10%, within the audio archive. Clearly, this is less than ideal for audio surveillance.
Accuracy, or to be more precise, lack of accuracy is also a big issue with many speech-to-text systems LVCSR dictionaries are limited to the number of words they contain making proper nouns particularly problematic to identify.
Phonetic search represents the next generation of speech analytics. The Aurix phonetic speech search engine is capable of monitoring up to 100% of recorded calls, enabling you to analyse a much wider sample of your recorded audio.
Unlike speech-to-text systems, our phonetics-based approach allows organisations to monitor audio recordings for non-standard words such as jargon, slang, and foreign words in native tongue. This can also be accomplished with real time audio streams. Users can move straight to the specific point in the recording where a word or phrase is used, bringing big improvements in productivity.
The Aurix phonetic audio search engine has an open vocabulary, to break free from dictionary constraints. Aurix indexes speech sounds into possible phonemes. Searches are conducted using phoneme strings derived from search words and phrases that the user inputs in text form. Multiple results are returned and ranked by confidence level.
Here are some of the many things you can do with Aurix phonetic audio search engine.
• Search your large audio and audio-visual archives for keywords and phrases ... including words, codenames or phrases previously unknown.
• Filter multiple feeds simultaneously in real-time ... security managers and analysts can now literally see or hear what others are saying about the topic(s) they are covering.
• Organisations can introduce up to 100% automated initial screen and filtering into their processes, hugely increasing the amount of intelligence released by focusing highly skilled human analysts on the most relevant material.
• Analyse large volumes of monitored speech in real-time for selected phrases or search packs. What are they saying? About what? And how often?
• Identify new words and phrases, now. Every day, new words and phrases are being coined; with Aurix you can stay ahead of the curve, not play catch-up. And, unlike speech-to-text based systems, you can search historical files without ever having identified that word or phrase previously.
To understand how Aurix solutions can improve your intelligence gathering efficiency, effectiveness and quality, please feel free to contact us.