Abstract: Techniques for personalized batch and streaming speech-to-text transcription of audio reduce the error rate of automatic speech recognition (ASR) systems in transcribing rare and out-of-vocabulary words. The techniques achieve personalization of connectionist temporal classification (CT) models by using adaptive boosting to perform biasing at the level of sub-words. In addition to boosting, the techniques encompass a phone alignment network to bias sub-word predictions towards rare long-tail words and out-of-vocabulary words. A technical benefit of the techniques is that the accuracy of speech-to-text transcription of rare and out-of-vocabulary words in a custom vocabulary by automatic speech recognition (ASR) system can be improved without having to train the ASR system on the custom vocabulary. Instead, the techniques allow the same ASR system trained on a base vocabulary to realize the accuracy improvements for different custom vocabularies spanning different domains.

Abstract: A face recognition neural network generates small dimensional facial feature vectors for faces in received images. A multi-task learning classifier network receives generated vectors and outputs classifications for the corresponding images. The classifications are built by respective sub-networks each comprising classification layers arranged in levels. The layers in at least some levels receive as input intermediate outputs from an immediately upstream classification layer in the same sub-network and in each other sub-network. After training, a resultant facial feature vector corresponding to an input image is received from the neural network. Resultant classifiers for the input image are received from the classifier network. A database is searched for a subset of reference facial feature vectors having associated classification features matching the resultant classifiers.

Abstract: A face recognition neural network generates small dimensional facial feature vectors for faces in received images. A multi-task learning classifier network receives generated vectors and outputs classifications for the corresponding images. The classifications are built by respective sub-networks each comprising classification layers arranged in levels. The layers in at least some levels receive as input intermediate outputs from an immediately upstream classification layer in the same sub-network and in each other sub-network. After training, a resultant facial feature vector corresponding to an input image is received from the neural network. Resultant classifiers for the input image are received from the classifier network. A database is searched for a subset of reference facial feature vectors having associated classification features matching the resultant classifiers.