Contact Me

I work for Voci, they produce high-performance speech recognition engine.   If you are looking for an all-round ASR solution for your product, I recommend you to talk with Voci's representatives.

Since I became the admin of AIDL, the PM requests are just overwhelming.   So I would suggest couple of ways you can follow me

What I would strongly recommend you not to do is to PM me.  As I said, the volume is overwhelming. (10 requests/day)

Donate to this Page:

7 thoughts on “Contact Me”

  1. Hi Arthur,

    I am a follower of your blog and I have been reading posts from your blog for quite a while now. The reason behind I'm writing this message, I am designing a speech recognition system for Bangla Language (my final year thesis), which is the native language of our country Bangladesh.

    There are huge resources on speech recognition for English, I would like to know, how I should approach to solve the problem. I won't be capable of making a huge dataset but I've talked to my supervisor and he said, overfitted model with small dataset would do just fine for now.

    I've taken CS224N - NLP using Deep Learning, CS231N - Computer Vision using DL, and the machine learning course from Andrew Ng. So I have some basic ideas on CNN, RNN and sort of stuffs.

    These courses are not directly targeted to solving problem like speech recognition, would it help to build one though? What's your thought? And what should I learn beside these? Would you please suggest me some resources and give a short guideline?

    I'm sure that's quite a lot of questions and you're a busy person, but if you can't answer all of them that's okay. Please take your time.

    Thank you.

    1. Hi Manash,

      Thanks for your message. Indeed, none of the deep learning courses you find on-line would teach you anything on current-day speech recognition. Due to IP reason (I worked for a ASR company), I also talked about speech recognition less. So I can only give you a cursory look of how you could learn it.

      ASR these days is very different from 5-6 years ago, and it's mostly deep-learning based. The key to understand the current day system is that senone-computation is no longer GMM-based, rather calculated through deep neural networks.

      So how you could learn it? In these days, the three great open sources recognizers survived are Kaldi, Sphinx and HTK. Starts with HTK, read through the manual and follow the exercises. Then try out Kaldi for accuracy, Sphinx for speed and APIs. They are all great recognizers, and forums have nice people to answer question. So try to see if you can gain more understanding. While these recognizers are still not the best in the planet, you will get reasonable results for a school project, a task or even for work.


  2. Hi Arthur

    My name is Vijay. I am an Acquisition Editor with Packt, a Developer Skills Platform.

    We are currently planning to develop a video courses on OpenCV 3 Computer Vision with Python Recipes & the other course is- Deep Learning projects with R and are looking for professionals who would like to author the course for Packt. I came across your Web-page and LinkedIn. Your knowledge and expertise in this technology cuts you out as a prime choice for authoring this course.

    I would love to get you on-board our project.

    Kindly, let me know your thoughts on this, post which we could discuss finer details such as course content & duration

    Looking forward to a positive response from you.

    1. Hey Vijay,

      Thanks for checking me out. Unfortunately, I am fairly busy in my work life so writing a book is really out of questions for me. In any case, hope things work out for you. Good Bye.


      1. Hi Arthur

        Thanks for your response.

        Actually, its not a book but a video course and its not a full time job. Almost all our authors are working professionals with day job. The process is such that you can manage it without disrupting your day today schedule. Do let me know and if this sounds good, then I can share the other details like process, duration and compensation.

  3. Hi Arthur,

    Hiton's Neural networks for machine learning seems to not be avaiable on Coursera any more. Can you recommend another way on which I can take it?

    Many Thanks

Leave a Reply

Your email address will not be published. Required fields are marked *