Research papers
Chihiro Arata, Shohei Taniguchi, Gouki Minegishi, Kan Hatakeyama-Sato
A variable-frame-rate speech codec with an exact utterance-level token budget and streaming decoding, enabling streaming text-to-speech generation.
An encoder-decoder codec language model for multilingual zero-shot TTS built on T5Gemma, with Progress-Monitoring RoPE for duration control.
Technical articles on Zenn (in Japanese)
Open-source projects on GitHub
A Flow Matching-based Text-to-Speech Model with Emoji-driven Style Control
Multilingual TTS model with voice cloning and duration control, based on T5Gemma encoder-decoder LLM
Inference server for MioTTS, a lightweight and fast LLM-based TTS model.
OpenAI Text-to-Speech API compatible server for Irodori-TTS
Models and datasets published on Hugging Face
Talks, lectures, and community work
Dec 17, 2025
DetailsAug 5, 2025
DetailsSep 2024 – Dec 2024
DetailsMay 2024 – Sep 2024
DetailsWork experience
Senior Research Engineer
Oct 2025 – Present
Research and development of AI models such as LLMs and TTS, from model training to deployment and operation.
LLM Engineer
Mar 2025 – Sep 2025
Developed LLMs for in-house products and built inference clusters.
Data Scientist
Apr 2023 – Feb 2025
Supported internal and client operations with generative AI.