Advancing the state-of-the-art in Automatic Speech Recognition (ASR) and Natural Language Processing (NLP) to create a delightful experience for customers (Swiss German speakers) of Swisscom Voice Assistant Box. Strengthening ASR capability to handle out-of-vocabulary (OOV) problem and dialectal variation in Swiss German by exploring and analyzing various linguistic and subword-unit-based approaches. Enhancing the recognition of Swiss German multi-dialect speech by transfer learning, unsupervised techniques and end-to-end translation to standard German. Boosting ASR performance of English code-switching in Swiss German by introducing novel techniques in both acoustic and language modeling.

Advanced the state-of-the-art speaker diarization system using end-to-end neural networks approach for speaker embedding. Achieved superior speaker diarization performance compared to the company's current solution. Implemented novel speaker recognition and diarization systems using deep learning technology for the final product. Designed and developed novel speech activity detection using neural network models. Explored novel techniques to enhance the single-channel speech intelligibility using neural networks. Demonstrated a high level of technical proficiency by participating in renowned challenges in speaker recognition and diarization including NIST SRE 2016, IBERSPEECH 2018, and the first DIHARD challenge 2018.