- Conducted research on voice cloning, speech synthesis, and few-shot learning to develop a TTS engine capable of synthesizing speech from limited audio data- Gained in-depth understanding of single-speaker and multi-speaker TTS models, their principles, and evolutionResearched end-to-end TTS models using transfer learning to clone user voices from as little as 162 seconds of audio- Analyzed and adjusted 2,800-speaker multi-speaker dataset- Trained and comparatively evaluated major TTS models using TensorFlow and PyTorch-based open-source frameworks, systemizing the model training process- Experienced multi-GPU parallel training in an Ubuntu-based GPU server and cloud environment

- Responsible for collecting and preprocessing text and image data.- Automated text data collection through web crawling.