Abstract: Embodiments included herein are directed towards a system and method for compressed domain language identification. Embodiments may include receiving a bitstream of a sequence of packets at one or more computing devices and classifying each packet into speech or non-speech based upon, at least in part, compressed domain voice activity detection (VAD). Embodiments may further include extracting a pseudo-cepstral representation from the speech detected packets and partially decoding without extracting a PCM format and generating a sequence of multi-frames, based upon, at least in part, the pseudo-cepstral representation. Embodiments may also include providing in real time the sequence of multi-frames to a deep neural network (DNN), wherein the DNN has been trained off-line for one or more desired target languages.

Abstract: Embodiments included herein are directed towards a system and method for compressed domain language identification. Embodiments may include receiving a bitstream of a sequence of packets at one or more computing devices and classifying each packet into speech or non-speech based upon, at least in part, compressed domain voice activity detection (VAD). Embodiments may further include extracting a pseudo-cepstral representation from the speech detected packets and partially decoding without extracting a PCM format and generating a sequence of multi-frames, based upon, at least in part, the pseudo-cepstral representation. Embodiments may also include providing in real time the sequence of multi-frames to a deep neural network (DNN), wherein the DNN has been trained off-line for one or more desired target languages.

Abstract: Typical textual prediction of voice data employs a predefined implementation arrangement of a single or multiple prediction sources. Using a predefined implementation arrangement of the prediction sources may not provide a good prediction performance in a consistent manner with variations in voice data quality. Prediction performance may be improved by employing adaptive textual prediction. According to at least one embodiment determining a configuration of a plurality of prediction sources, used for textual interpretation of the voice data, is determined based at least in part on one or more features associated with the voice data or one or more a-priori interpretations of the voice data. A textual output prediction of the voice data is then generated using the plurality of prediction sources according to the determined configuration. Employing an adaptive configuration of the text prediction sources facilitates providing more accurate text transcripts of the voice data.

Abstract: Typical textual prediction of voice data employs a predefined implementation arrangement of a single or multiple prediction sources. Using a predefined implementation arrangement of the prediction sources may not provide a good prediction performance in a consistent manner with variations in voice data quality. Prediction performance may be improved by employing adaptive textual prediction. According to at least one embodiment determining a configuration of a plurality of prediction sources, used for textual interpretation of the voice data, is determined based at least in part on one or more features associated with the voice data or one or more a-priori interpretations of the voice data. A textual output prediction of the voice data is then generated using the plurality of prediction sources according to the determined configuration. Employing an adaptive configuration of the text prediction sources facilitates providing more accurate text transcripts of the voice data.