This project serves as a proof-of-concept for integrating voice recognition into programming, with a focus on overcoming a notable challenge in NLP: accommodating various accents in voice recognition.
Leveraging Transfer Learning, it showcases the ability to identify new limited-keywords with minimal fine-tuning, while also accommodating the accent inherent when training new words.
TensorFlow • NLP • Transfer Learning • JavaScript
Innovative Approach: Applying Machine Learning concepts to address the practical challenge of enhancing speech recognition in the NLP sphere.
Transfer Learning: Skillfully implementing transfer learning, which significantly reduces the time and resources needed to develop accurate Machine Learning models for speech recognition.
Future Work: This proof-of-concept is an attempt at factoring in accents with voice recognition. A fully-built product could look like asking the user directly for 5 voice samples of a certain keyword, or a single long sample of a paragraph.
At the highest level, a single model would be deployed that is robust in understanding English that is heavily accented, such as countries where English is not the primary language.
Demonstrating specific keyword recognition by programming and navigating the application menu.
