Neural Machine Translation, Transliteration and Voice assistance Chatbot

SIH 2019 for Kotak Mahindra Bank

The objective is to enable Kotak Mahindra bank’s Keya chatbot support voice channel and accepts multilingual input. The model includes natural language understanding, speech recognition, machine translation. It handles various services provided by bank. It takes human voice as input and processes it and provides output as both text and voice. The user can select any language which he/she is comfortable with and can get any information regarding bank services.

  • IBM-Discovery — Helps in conversation by taking unstructured data.
  • IBM-Assistant — It has good support for large business enterprises.
  • IBM-Natural Language understanding .
  • IBM-Tone Analyzer — Used for sentimental analysis by converting voice into text and then analyze it.
  • Google Translator.
  • Google Speech to text.
  • No form filling.
  • Hard code is trained for particular words. Example: Idiot.
  • Can process only English Language.
  • No feature to locate nearby ATM’s.Automatic form filling by speech.
  • It understands Human Emotions.
  • Has multilingual support.
  • Nearby Bank or ATM locator.
  • Complaints Analysis.

Using APIs suffice the job to get done. If you are interested please follow along to get insights to build your own network.

Alternative idea

Sequence2Sequence model

Here we use Encoder and decoder model for machine translation.