Get to Know Your Fellow Researchers: Jim Glass

February, 2017

What sparked your interest in speech and language processing?

I have always been interested in artificial intelligence, especially as it relates to perception. As I learned more about signal processing as an undergrad, the problem of automatically processing speech appealed to me, and was something I pursued in my graduate research. Our field lies at the intersection of signal and symbolic processing, and incorporates ideas from electrical engineering, computer science, linguistics, brain and cognitive science, and others. It is at a great crossroads. Besides, there’s something about digging out symbolic-level information from a noisy signal that I find fascinating.

How do you think speech and language processing is changing the society for the new generation?

Over the last five years or so we have crossed a tipping point whereby speech and language processing have become pervasive in our society. It started on our smartphones, but has moved into the car and home, and is on its way to becoming ubiquitous. What seemed like science fiction not too long ago has become reality, and it’s just assumed, moving forward, that our devices will be speech enabled.

What is your holy grail in speech and language processing? When will we achieve it?

My holy grail is automatic language acquisition - the ability of a machine to learn a new language in much less supervised environments than are currently employed to train speech recognizers. There is a practical reason for this, in that most languages of the 7,000 or so spoken languages in the world have received no attention from ASR due to lack of resources. My holy grail challenge would be be able to plunk a robot down somewhere in the world and have it interact with humans and learn a new language on its own. The follow on would be to learn how to translate between these languages automatically. I don’t know when we will achieve this, but we are moving inexorably in that direction.

Do you have any specific advice for students, junior faculty or others early in their careers?

I encourage students not to follow the well-worn path, but to look for new frontiers that people haven’t thought about yet. There is a tendency to look at research that is being performed, and find some method that incrementally improves performance. This is especially true when there are well-established experimental paradigms with relevant data. It’s harder to think of the important problems that people aren’t working on yet, but I find those are often the most fun things to explore. There’s nothing like making fresh tracks!

What development in the field has most surprised you? Was there a hard problem that turned out to be easy? An easy problem that proved surprisingly difficult?

We are awash in the deep learning tsunami right now, so I will say I have been impressed how these methods seem to be finding their way into all aspects of speech and language processing (as well as all the other machine learning problems). It’s an exciting time to be working in this area, and so many opportunities are presenting themselves, and the field is moving so quickly!

Which of your publications is your favorite? Why?

My favorite is perhaps an article on unsupervised speech pattern discovery that was based on the doctoral work of my former student Alex Park. It developed a method for finding clusters of word-like units based on repeating sound patterns, without knowing anything a priori about what content, or even language, was being spoken. It by-passed conventional speech recognition to look back at the audio signal directly. It is a nice example of research that opens up new frontiers. Now there is much more activity on the so called zero-resource problem - what can you learn form large quantities of unannotated speech data.

Publications & Resources

Conferences & Events

Professional Development

Community & Involvement

About IEEE SPS

For Volunteers