ALOK PARLIKAR

A lot of progress has been made in the field of speech synthesis in the last few years. Today's synthetic voices sound natural and are quite intelligible. A lot of this progress can be attributed to "resource development" in languages. Large databases of speech, linguistic analysis tools, pronunciation dictionaries, text processing rules, etc. can help make synthetic speech significantly better. However, this privileged "resource development" is only enjoyed by a few languages of the world. Text to speech (TTS) has become a common utility today. We now hear synthetic voices at airports, railway stations, in our GPS navigation devices, and even in toys. This phenomenon has led to an increasing demand for deploying TTS in localized languages. Most of these languages belong to the "low-resource" category. Although linguistic resource development work is happening at a fast pace for Indian languages, for many tasks, they still fall in the low resource category. In this talk, we will take a look at some challenges in building TTS systems for low resource languages. We will discuss in detail the challenge of inserting pauses between words (phrasing) in synthetic speech and look at novel data driven techniques that also work in low resource scenarios.

Biography:

Dr Alok Parlikar graduated from the Premier Indian Institute of Information Technology, Allahabad with a gold medal. Subsequently he joined EPFL University , Switzerland for post graduate research before joining the computer science department of Carnegie Mellon University, USA from where he received his MS and Ph.D degree. He aslo worked at Amazon, USA for one year.