職務内容

DubGuild is developing the "World’s Best AI Dubbing Technology" with the goal of enabling content from anywhere to be delivered across the globe without losing its original appeal.
Our products provide AI dubbing technologies that capture the full expressive range required by voice actors and translators, as well as user-friendly interfaces for applying them.

Why are we developing speech language foundation models?
Large language models (LLMs) have attracted significant attention in recent years for their text-based conversational capabilities, and have greatly improved human productivity. However, since much of human communication is inherently conducted through speech, there is a growing need to move toward speech-based models.

The most straightforward way to achieve this is to use a pipeline consisting of automatic speech recognition (ASR) + LLM + text-to-speech (TTS). However, this approach has limitations in emotional expression and real-time responsiveness, which have become barriers to realizing and widely deploying natural voice interactions.

This is why we are developing speech language foundation models: models that can understand, reason with, and generate speech directly, without treating speech merely as an input/output interface for text. This is a technologically challenging field that is still in its early stages globally. DubGuild is taking on this challenge from Japan, aiming to build a world-class speech AI foundation model.

For more details, please refer to our tech blog:
https://blog.dubguild.com/melte/llm-tts-scaling/

After joining, you will primarily belong to one of the following four teams. We also welcome cross-functional contributions across teams.

(A) HPC: Accelerating preprocessing of speech and language data, and executing large-scale pretraining
(B) Storage / Network: Stabilizing communication for large-scale datasets and building data infrastructure
(C) Model Serving / API Infrastructure: Building inference infrastructure and improving stability and responsiveness
(D) ML / NLP / Speech: Preprocessing, tokenization, SFT, RL, and model evaluation using domain expertise

If you work with full time job, we can fully support your VISA. We have some experience of it.

---

採用の流れ

応募→チャット欄で議論→面接→採択

---