Spoken Term Detection (Audio Retrieval) [Project Report] [Presentation]

Research Intern at Indian Institute of Technology, Kanpur, India.
  • Aimed at solving the task of Spoken Term Detection (STD) to retrieve queried speech files in an audio database.
  • Implemented three different approaches to STD for query localization, classification and location suggestion in a database.
  • Analyzed an optimal combination of the above, in order to work towards building a language-agnostic system.

Statistical Machine Translation [Presentation]

Associated with Language Technologies Research Centre, International Institute of Information Technology, Hyderabad, India.
  • Aimed at investigating the effects of different evaluation metrics during the tuning phrase of an SMT model.
  • Compared the results of models tuned on BLEU, chrF, TER, WER and PER on a subset of the IndicWAT corpus.

Neural Models for Natural Language Processing

Associated with IvLabs, Visvesvaraya National Institute of Technology, India.

Neural Machine Translation [GitHub]

  • Studied papers presenting novel architectures for NMT.
  • Implemented Encoder-Decoder architectures in PyTorch using the Multi30k Dataset for German-English.
  • Low-resource NMT (Literature Review): [Presentation] [Notes]

Sentiment Analysis [GitHub]

  • Aimed at the automatic determination of polarity in text.
  • Compared the results of different architectures such as LSTM, FastText and BERT on the IMDb dataset.

Automatic Speech Recognition [GitHub]

  • Aimed at developing an end-to-end ASR system.
  • Deployed Deep Self-Attention Networks [arxiv] in Keras using the LJ Speech Dataset.

Name Generation [GitHub]

  • Generated dinosaur names by developing a character-level Language Model using PyTorch.
  • Compared the results of Vanilla RNN, LSTM and GRU.

Notes