Few-shot meta multilabel classifier for low resource accented code-switched speech

Low resource end to end automatic speech recognition (ASR) system development is challenging due to the scarcity of annotated speech corpus. Meta learning is being widely used as a paradigm for few shots learning for fast adaptation on unseen tasks as well as for low resource downstream tasks. Recent studies prove the feature reuse property of meta learning while training on new tasks. We explore multi label classification using low resource conversational Malayalam-English code-switched speech. Code-switched speech has majorly inter-sentential, intra-sentential and intra-word code-switching with English. Apart from English there is presence of other languages in minor quantities. Conversational Malayalam have multiple dialects and meta sounds as well. Annotating speech data in low-resource code-switched settings is challenging due to limited access to resources such as trained bilingual transcriptionists and pre-labeled data for validation. In this work we present the annotation of low resource dialectal code-switched conversational speech as a few shot multilabel classification problem. We present a unified classifier chain meta training algorithm using feature reuse property of Almost no Inner Loop (ANIL). The experimental results on classification accuracy of multilabel classification in few shots setting for Malayalam-English code-switched speech with meta feature reuse was also presented.