Visual Bank Expands “Qlean Dataset” to Support Large-Scale Japanese Speech Foundation Models

Delivering 100,000+ hours of rights-cleared Japanese audio, including regional dialects and culturally contextualized speech essential for commercial AI development.

Visual Bank Inc. (CEO: Saneyuki Nagai), through its subsidiary amanaimages Inc., one of the largest digital asset providers for the marketing and advertising industry in Japan with over 40 years of history, today announced the expansion of its Qlean Dataset, a premium AI training data solution designed for developers building high-performance Japanese speech foundation models.

This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20260401752248/en/

Visual Bank Group, leveraging over 40 years of expertise through amanaimages Inc., expands Qlean Dataset, delivering high quality, rights cleared Japanese language corpora, including 100,000+ hours of commercially usable audio.

A new development within the Qlean Dataset division, which focuses on providing datasets for institutions engaged in research and development, with rights cleared for AI training and large-scale data applications, has positioned the company as a leading provider of Japanese language data infrastructure, particularly for structured Japanese speech corpora organized by speaker configuration and thematic domain.

Key Features for AI Developers

• Rights-Cleared Data for Commercial Use
  All datasets are fully rights-cleared for commercial use and aligned with global compliance standards such as GDPR and CCPA.
• High-Fidelity Audio Assets
  Recordings at 48kHz/16bit or higher capture both studio-quality speech and acoustic environments.
• Expert Human Annotation
  Native-level transcripts and structured metadata provide training-ready datasets for AI development.
• Safety and Moderation Datasets
  Datasets support detection of harmful language, including hate speech and abusive prompts.
• Japanese Evaluation Datasets
  Evaluation datasets are aligned with international benchmarks such as MMSU to measure reasoning and linguistic nuance in Japanese.
• Japan-Specific Acoustic Environments
  Japan-specific audio, including traditional instruments, shrines, and urban environments, supports multimodal and spatial AI.

These datasets are available through AI Data Recipe, a flexible offering that provides both ready-to-use datasets and custom data production, including speaker casting, recording, and annotation tailored to specific model architectures and development needs.

“As demand for culturally contextualized foundation models grows, high-quality, legally compliant Japanese training data is becoming increasingly critical,” said Saneyuki Nagai, CEO of Visual Bank Inc. “Visual Bank is committed to bridging the gap between raw content and production-ready AI systems through rigorous data preparation and engineering.”

AI Data Recipe
https://qleandataset.visual-bank.co.jp/en/lineup

Japanese Language Corpora
https://qleandataset.visual-bank.co.jp/en/products/japanese-language-corpora

View source version on businesswire.com: https://www.businesswire.com/news/home/20260401752248/en/