Text Analytics from Audio

Audio transcription, translation and Natural Language Processing (NLP) capabilities are required to automate the extraction of actionable information from foreign language audio files. The problem is that these language processing components are traditionally stand-alone and pipelined together producing accumulative growing word error rates, significantly degrading the level of trust. To overcome the shortfalls of a multi-staged pipelined approach we propose a two stage approach comprised of Automated Speech Recognition (ASR) transcription, then a Foreign Language Analysis (FLA) stage that goes directly from foreign language text to English concept encoded in FrameNet semantic frames. Given the innovation and newness of the FLA approach our proposal focuses on this stage, leveraging internal stages of SYSTRAN"s translation capabilities as commercial device technologies evolve, more and better ASR solutions will be inherent available such as the iPhone Siri and Android Google Voice. The ability to have a single component or app to go directly from these device ASRs to semantic frames will result high levels of trust while easing integration with applications that need to consume actionable information. The plan is to produce 200 Farsi sentences with various tones and noise, use a commercial Farsi ASR and co-develop a Farsi FLA with SYSTRAN.