Abstract: Techniques for performing spoken language understanding (SLU) processing on a device are described. Example embodiments involve a device determining whether a spoken input corresponds to a supported spoken input class, a supported spoken input with dynamic content class or an unsupported spoken input class. For a spoken input corresponding to the supported spoken input with dynamic content class, the device may determine an entity corresponding to the spoken input from a set of entities, which may be determined based on device context data and/or user profile data. For a spoken input corresponding to the supported spoken input class, the device may determine an intent and entity using stored data. For a spoken input corresponding to the unsupported spoken input class, the device may send the audio data to a system for processing.

Abstract: Techniques for biasing for entities during automatic speech recognition (ASR) processing are described. In some embodiments, a system implements a gating component that is configured to switch on and off entity biasing on an audio frame basis when processing a spoken input. The gating component processes an audio frame to determine whether the audio frame likely includes a representation of a custom entity. Based on the determination, a biasing component, which is configured to generate entity embeddings, may be turned on or off. In this manner, entity biasing does not run on every audio frame, but only on the audio frames where it can be helpful in increasing ASR accuracy.

Abstract: A processor-implemented method for spatial sound characterization is described. In one implementation, each of a plurality of source signals detected by a plurality of sensing devices, is segmented into a plurality of time frames. For each time frame, time-frequency transform of the source signals is derived, an estimated number of sources and at least one estimated direction of arrival corresponding to each of the source signals is obtained. Further, source signals are extracted by spatial separation based at least on the estimated directions of arrival and the estimated number of sources, and separated source signals are processed to yield a reference signal and side information.

Abstract: A processor-implemented method for spatial sound localization and isolation is described. The method includes segmenting, via a processor, each of a plurality of source signals detected by a plurality of sensors, into a plurality of time frames. For each time frame, the method further includes obtaining, via a processor, a plurality of direction of arrival (DOA) estimates from the plurality of sensors, discretizing an area of interest into a plurality of grid points, calculating, via the processor, DOA at each of grid points, comparing, via the processor, the DOA estimates with the computed DOAs. If the number of sources is more than 1, the method includes obtaining via the processor, a plurality of combinations of DOA estimates, from amongst the plurality of combinations, estimating, via the processor, one or more initial candidate locations corresponding to each of the combinations, selecting location of the sources from amongst the initial candidate locations.

Abstract: A processor-implemented method for spatial sound localization and isolation is described. The method includes segmenting, via a processor, each of a plurality of source signals detected by a plurality of sensors, into a plurality of time frames. For each time frame, the method further includes obtaining, via a processor, a plurality of direction of arrival (DOA) estimates from the plurality of sensors, discretizing an area of interest into a plurality of grid points, calculating, via the processor, DOA at each of grid points, comparing, via the processor, the DOA estimates with the computed DOAs. If the number of sources is more than 1, the method includes obtaining via the processor, a plurality of combinations of DOA estimates, from amongst the plurality of combinations, estimating, via the processor, one or more initial candidate locations corresponding to each of the combinations, selecting location of the sources from amongst the initial candidate locations.