We dont realise the extent of unstructured data we have generated from recordings of our office meetings, conversations with doctors or phone calls with customer service professionals. This data contains lots of valuable information that companies can utilize to make the right decisions or for healthcare professionals to develop better interventional strategies.
Prachi Singh, our next pathbreaker, applies machine learning and signal processing techniques to solve problems in the area of speaker diarization, which is basically partitioning a multi-talker audio recording into different segments based on speakers present in it.
Prachi talks to Shyam Krishnamurthy from The Interview Portal about leveraging her interest in mathematics and background in Electrical Engineering to take up Speech Processing, with a blend of signal processing and machine learning.
For students, lots of companies have realised the value of unstructured data. A great research career that can lead you to the industry to address big problems.
Prachi, tell us about your background?
I am Prachi Singh. I grew up in Gondia, a district in Maharashtra. It is not a very big place, but a small market without fancy malls and less traffic to deal with. My father is a railway engineer and mother is a homemaker. My parents, hailing from a family of farmers, knew all the hardships in the life of a farmer and the importance of education in life. They always encouraged and supported me and my brother to study. Still I was not under any pressure whatsoever . I used to participate in other activities like dance, sports etc. I have enjoyed mathematics and science since my school days. I never got bored solving mathematical problems. I was not sure of my career but I never wanted to leave maths behind in whatever I did, which led me to do engineering.
What did you do for graduation/post graduation?
I did my B.Tech from College of Engineering, Pune in Electronics & Telecommunication in 2015. I received a scholarship from Dhirubhai Ambani Foundation. Currently, I am pursuing my PhD from Indian Institute of Science, Bangalore in the field of speech processing. I am supported by a MHRD scholarship.
What made you choose such an offbeat, unconventional and interesting career?
Since my school days, I enjoyed mathematics and science. My father is a railway engineer. I was fascinated by seeing many electronics devices at home and him repairing gadgets at home. This inspired me to pursue engineering. There I enjoyed learning subjects related to algorithms and signal processing apart from mathematics. I wanted to go deep into these subjects. My uncle also inspired me to study further and guided me along the way. I wanted to do a PhD someday in the future, but didn’t know that I would get that opportunity so early.
Tell us about your career path
I believe that we need not be always sure of our final goal. It is important to have short term goals which will lead to our final goal. In my B.Tech, I did projects involving signal processing like Optical Character Recognition (OCR) and ECG signal analysis for drowsiness detection. After completing my B.Tech in 2015, I worked in Fiat Chrysler and Automobiles (FCA) for two years. It is an automobile company which has a separate department for electrical engineering. There I came to know how the backend of all the features we see in our cars works and what makes each feature look different from others. I was amazed at seeing vehicles which looked like an assembly of mechanical parts, and meshes of electric wires inside. These wires carry signals that enable communication among different parts. I learnt about CAN bus protocol and worked with instrument panels which display warnings in vehicles. Meanwhile I developed an interest in machine learning. It fascinated me that we could use data and samples to predict trends or classify them into one of the given classes.
I wanted to learn more about all these subjects. I also wanted to explore and work on something new. This encouraged me to study further. It was not so easy but not impossible either. People said it would be difficult to get back to studies once you start working. But nothing stopped me, as I was already determined. I always dreamed of studying in IISc because of its environment, literally and figuratively. When I got admission into IISc, I had to choose my advisor and area of research. I had already worked in Image processing earlier and at IISc, I found other options like speech processing. Speech processing has, of late, been getting more attention, especially in industries, driven by the need to make better Graphical User Interfaces (GUIs). All companies are working to integrate features with human voice for a richer human experience. I also visited labs working in this area. One such lab was the Learning and Extraction of Acoustic Patterns (LEAP) lab headed by Dr. Sriram Ganapathy. They were working on various applications of speech with a blend of signal processing and machine learning. I was very excited about the work and immediately contacted Dr. Sriram Ganapathy, asked for more details and requested to join the lab. Since then, I am a part of the lab, attending seminars, reading literature, trying different experiments and publishing papers. It is an ongoing process.
How did you get your first break?
I wrote the Graduate Aptitude Test in Engineering (GATE) exam with the aim of doing M.Tech. I applied to various IITs and IISc. I got an interview call from IISc for M.Tech (Research) in the Electrical Engineering department. When I was filling out the forms for my areas of interest at IISc, there was one question asking me if I wanted to be considered for a direct PhD. I never knew about the option of doing a direct PhD without completing M.Tech. I consulted my uncle and he said that it was a very good opportunity and not to let it go. So I opted for it. My interview went well, and the panel was very supportive. After a few weeks, I received a letter saying that I have been selected for direct PhD . Finally I joined IISc in July 2017, as a PhD student.
What were the challenges you faced? How did you address them?
Challenge 1: First challenge I faced was expected, since I am a direct PhD scholar, with no master’s degree and no previous exposure to the research environment. Till graduation, I always studied fixed subjects with a well-defined syllabus. We gave exams and we were free. Research is not like that. You need to figure out the area you want to work on, the problems in that area you want to address, and the knowledge you need to explore that problem. Initially I was a bit scared, finding myself undeserving in this field. But my advisor, Dr. Sriram, helped me sort out my way. He suggested subjects to study, papers to read and the skills to develop which gave me the confidence to proceed.
Challenge 2: Another obstacle was the unexpected outcomes. In research we can never know if we are going in the right direction, especially in the beginning, and that’s why we are researching. In many cases, days will go without getting the expected results. Some days, we will get results immediately. A researcher needs to have lots of patience and strength to stay focused. It took me some time to adapt to it. I used to feel frustrated and hopeless when things were not moving ahead, because in B.Tech, we always did those things which have already worked before and proven, but here it was different. Moreover, the world is very competitive, and every minute you will find someone else publishing a paper which creates a lot of tension. But then I realized that I needed to understand and learn from my mistakes. There is nothing wrong in not getting expected results but the idea is to understand why and how to take the next step. PhD is not just a degree, it teaches life lessons.
Where do you work now?
I am a PhD student in Learning and Extraction of Acoustic Patterns (LEAP) Lab, Electrical Engineering, Indian Institute of Science, Bangalore working in the area of speech processing.
My research involves application of machine learning and signal processing in solving problems in speech. I am working in the area of speaker diarization. This means partitioning a multi-talker audio recording into different segments based on talkers or speakers present in it. This problem, although it looks simple, involves a lot of challenges. The recordings which we have are taken from the real world, having different forms of noise in the background, speaking rate is high with rapid speaker turns and many other such issues.
What skills are needed for a job? How did you acquire the skills?
In my work, I need skills in programming (mostly python), Data structures, Machine Learning algorithms, concepts from linear algebra, probability, optimization, speech processing etc. To acquire these skills, I took courses and did projects in relevant fields. The best thing about IISc is we can take courses from any department, for example, students from biology can take courses from computer science, and neuroscience students can take courses from mechanical. This helps researchers and students work in multidisciplinary areas and increase the horizon of our knowledge. We also participate in various online challenges, compete with world universities and industries to know where our work stands.
What is a typical day like?
My typical day in the lab starts with making notes of what to finish by evening. I experiment with various datasets, analyzing my models. I discuss my results with my advisor, every alternate day. To keep myself updated, I tend to read the latest publications in my field. IISc also provides facilities for extra-curricular activities like badminton, swimming. It depends on you how you plan your schedule.
What is it you love about this job?
The best thing I love about what I do is, I have the liberty to plan my work. Although there are submission deadlines, you can make your own schedule to achieve those. I am very fortunate to have a healthy work environment with very supporting and ambitious colleagues, working day and night, who inspire me to push further. A PhD provides exposure to many areas of research, meet prominent people across the world, discuss ideas, and collaborate with industries.
How does your work benefit society?
My work is useful in areas where we collect speech data to analyze various things. It can be meetings, doctor-patient conversations, telephone recording in call centers, electronic devices like voice assistants, voice password and so on. In such cases, listening to the recordings for performing tasks like speech to text conversion, speaker identification, speaker segmentation and noise removal are very tedious and time consuming tasks, which makes it necessary to build machines to automate the process. The end-goal is always domain specific. We can use the output or results to transcribe meetings, track patient behavior, understand customer satisfaction levels, spoof detection, answer a query and so on.
Your advice to students based on your experience?
Many people are scared to do a PhD because of some prior beliefs. I want to explain some of these to encourage students to pursue PhD or at least consider it as an option.
- PhD involves lots of studying : Yes it does, but it is unlike your school where you studied everything which was needed. Here you select only those subjects, ideas, theories which you actually are curious about and want to know in depth.
- It needs lots of hard work : Of course it does. But whatever career you choose, you need to work hard to excel in that. PhD opens the door to explore things which nobody knows about and understands why, how or what happened. You have the freedom to do all the experiments which can help you find these answers.
- It takes a lot of time to complete : It may or may not. It depends on your area and efforts. You cannot expect wonders to happen overnight. If you can spend your 30-40 years working in a company, can’t you spare 5-6 years to explore something new.
- What after PhD ? I don’t want to teach : Teaching is not the only option. There are students who join research labs in Google, Microsoft, Facebook, as well as your dream companies, ISRO, NASA, Science laboratories etc. Doors will open to many possibilities.
- Is it difficult for women ? People will always say things become difficult once you get married and have kids but I want to say that there are many women doing PhD who are married and have kids, and they are able to do very well with family and advisor support. It is easier than working in a company where you have to comply with fixed working hours in a week, but here you have the flexibility to work according to your time. Being a woman, I want to encourage all school and college girls not to be afraid. Once you are determined everything is possible.
- Masters in the US is a better option : Generally students tend to go to the US for masters. But I think India also has many good institutes on par with world standards. Before going abroad check the options in India, because apart from good education, it is way cheaper and has all the facilities like hostel mess, which is not easily available abroad.
I encourage students to acquire multiple skills and be aware and updated about the latest trends in various fields, try to apply your knowledge through small projects, internships and project assistantships in universities. Keep solving problems apart from school curriculum.
Currently, my focus is to make my PhD more impactful and I am still in the beginning. My aim is to work with professionals and get enough exposure to be able to start some work of my own.