Audio Tech leader. Researching, developing and deploying production-level Text-to-speech, Voice cloning, voice conversion services for interactive agents. - Parallel Text to speech- Autoregressive TTS- Zero-shot, Few-shot Voice cloning- Emotional voice synthesis- Singing voice synthesis- Speech enhancement

Production: Flax scanner: a system that accurately extracts information from visually rich document such as application forms, business documents, certificates, invoices, license ... Some used techniques: Image semantic segmentation, Object detection, Optical character recognition (OCR), Graph convolution neural network, Document's chargrid segmentation ...-Participate in software development process.-Develop and fine-tune deep learning models to deliver production-ready AI solution.-Take the role Project Technical Leader for 1 project. -Conduct applied researches: Document object detection, Japanese handwriting character generation, Confidence score for document information extraction system.Production: Aurora Clipper: a system that finds important information from long sequence digital texts like E-mails or contracts, business documents, ...In the scope of this product, I've conducted some related applied researches: Information extraction with human feedback (applying Reinforcement learning techniques to text summarization, document retrieval and question answering models).Internal product: Data managment tool. Internal data managment and data annotation tool. My role is building some parts of backend API, serving Deep Learning model for auto data labeling.Used techniques: Django, Django REST framework, Docker, Celery, Ray Serve ...

BU Data Scientist, Data Analytic Center, Viettel Telecom.Apply machine learning algorithms in big data processing, focus on telecom industry data and predictive analysis.- Techniques: Collaborative filtering, content-based filtering, Decision Tree, ...- Tools: Python, Spark, SQL ...

Worked in Deep learning and image processing.Project: Artistic Style Transfer on Mobile DevicesApply deep learning/computer vision solutions to the style-transfer problem. (like Prisma app: transfer the realistic photo of the user to an artist painting).Deploy the deep learning model to server.Techniques: Fast style-transfer, patch-based style-transfer, image semantic segmentation, face detection.- Techniques: Fast style-transfer, Fully Convolutional Networks for Semantic Segmentation, ...- Tools: Caffe, Keras, Tensorflow, OpenCV, C/C++, ...