AALF: Adjustable Adaptive Language speech Filter

The ability to efficiently process intelligence data is critical to fighting the global war on terror. A major source of this data is recorded speech. Therefore, the performance of speech processing applications (speech recognition and speaker identification applications, for example) depends on pre-processing to find the regions in the recording where speech is present. Although current methods have had success isolating human speech from interfering background noise, these methods have not addressed non-language speech sounds (NLSS), i.e., sighs, coughing, and backchannel sounds (“un-huh”, “hmm”, etc.). 21st Century Technologies (21CT) proposes Adjustable Adaptive Language speech Filter (AALF) as a robust NLSS filtering system that combines a signal processing feature extraction component, an adaptive machine learning mechanism that provides robustness to various input conditions, and a post-processing component that includes a user-specified detection threshold to tune AALF’s performance for specific speech processing applications that follow. AALF will improve the generation of actionable intelligence by decreasing the cost of creating speech training databases and improving the performance of speech processing applications.