Acoustics Based Intent Recognition Using Discovered Phonetic Units for Low Resource Languages

With recent advancements in language technologies, humans are now speaking to\ndevices. Increasing the reach of spoken language technologies requires building\nsystems in local languages. A major bottleneck here are the underlying\ndata-intensive parts that make up such systems, including automatic speech\nrecognition (ASR) systems that require large amounts of labelled data. With the\naim of aiding development of spoken dialog systems in low resourced languages,\nwe propose a novel acoustics based intent recognition system that uses\ndiscovered phonetic units for intent classification. The system is made up of\ntwo blocks - the first block is a universal phone recognition system that\ngenerates a transcript of discovered phonetic units for the input audio, and\nthe second block performs intent classification from the generated phonetic\ntranscripts. We propose a CNN+LSTM based architecture and present results for\ntwo languages families - Indic languages and Romance languages, for two\ndifferent intent recognition tasks. We also perform multilingual training of\nour intent classifier and show improved cross-lingual transfer and zero-shot\nperformance on an unknown language within the same language family.\n