Introducing TWIML & AI’s newest intern!
Hi everyone! I’m Malaika Charrington, TWIML & AI’s newest intern. I’m a recent high school grad and will be attending Denison University in the fall with an intended major in computer science and psychology. This summer I’ll be working on a project in which I evaluate the efficacy of a variety of automated transcription services in order to develop an efficient transcription process for podcast shows, and use natural language processing techniques to automatically tag podcasts to improve their discoverability.
When my dad offered me the position, I was wary, as I haven’t worked much with machine learning or AI content (and to be completely honest, I wasn’t sure that I even wanted a summer job), but in the two weeks since I started I’ve been pleasantly surprised! I’ve begun learning Python, which is very different from the C++ I learned in school, I’ve done some research and analysis for my project, and I’ve given a short, successful presentation on my findings! I’m excited to discover new things about machine learning and artificial intelligence and broaden my skill set.
About My Project
Phase 1 of my project is dedicated to identifying and researching different transcription services and their unique features. The services that I have identified and researched for this phase were Deepgram Brain, Speechmatics, Microsoft Speech-to-text, Google Speech-to-text, Amazon Speech-to-text, IBM Watson Speech-to-text, VoiceBase Speech-to-text, Trint, Temi, and Sonix.
In phase 2, we will prepare test audio files consisting of several speakers with different accents reading from a general text, as well as several speakers reading technical text, and use these to test out each of the transcription services for accuracy and ease of use, and correcting them to create a source of truth, and narrow our list to five or six services. We’ll then continue testing against real podcast audio, and further compare the performance of the different services.
The next step, phase 3, will consist of writing a Python script to quickly and easily quantify the accuracy of each service, following which, I will write a report on my findings. Then, for phase 4, I will attempt to minimize inaccuracies by taking advantage of the chosen services’ different features. In phase 5, I will begin researching different services that perform automated semantic tagging and I will experiment with different tagging approaches using NLP. Next, for phase 6, I will be adding this semantic tagging to the website, to allow users to easily identify podcasts about the subjects that interest them the most. Finally, I will give a final presentation and write up covering the full project, and showing the final product.
I’ll be checking in here frequently with updates about my project. This internship has already begun to help me refine my research and writing skills and I’m so eager to take on the challenges ahead and see the other ways that I grow over the next few months!
