Dataocean AI

04/15/2026

🚀 AI EXPO TOKYO 2026 | Day 1 Kickoff! 🇯🇵
📣 We are thrilled to kick off Day 1 of AI EXPO TOKYO 2026 at Tokyo Big Sight (West Hall)!
📍 Visit us at Booth #1-31A to explore our flagship offerings:
✅ High-precision Japanese & multilingual datasets
✅ Robust data solutions powering LLMs, speech recognition, synthesis & Generative AI
✅ Customized data strategies for enterprise AI applications
A heartfelt thank you to all who’ve visited our booth today. We’re excited to continue connecting with industry leaders and innovators!
Our data experts are on-site to discuss how our data solutions can accelerate your AI projects. Don’t hesitate to schedule a meeting or stop by for a chat.
Looking forward to meaningful collaborations over the next two days!

12/26/2025

🚀 Multimodal Dataset Release! 📢
We’re pleased to introduce two newly released multimodal datasets, designed to support multimodal foundation models, vision-language understanding, aesthetic evaluation, image generation, and creative AI applications.
What’s new:
🎥 Human–Object Interaction Action Dataset
Video-based data capturing human–object interactions across real-world scenes, with spoken descriptions to enhance action and semantic understanding.
🖼️ Aesthetic Composition Training Dataset
High-quality images with professional composition labels, supporting aesthetic evaluation, image generation, and visual enhancement models.

📩 For samples or collaboration, feel free to reach out.
🔥 Explore the datasets: https://dataoceanai.com/datasets/cv/aesthetic-composition-training-corpus/
https://dataoceanai.com/datasets/multimodal/handheld-object-portrait-corpus/

12/16/2025

🚀Interspeech 2026 AECC — Now Open!!!
The 2nd Audio Encoder Capability Challenge (AECC), co-organized by Xiaomi, University of Surrey, Tsinghua University, and Dataocean AI, is now officially open for registration.
✔ Participants only need to submit a pre-trained audio encoder; all downstream training and evaluation are handled by the organizers.
✔ An open-source evaluation system, XARES-LLM, is provided by the organizers.
✔ Two tracks are available:
• Track A focuses on LALM performance on traditional classification tasks.
• Track B emphasizes higher-level understanding and expressive generation capabilities.
✔ DataoceanAI provides an auxiliary dataset, constructed from eight commercial-grade datasets to support the competition.
📅 Registration closes Jan. 25, 2026, 11:59 PM AoE
📩 Contact: [email protected]
For more details about the challenge, please visit the official page: https://dataoceanai.github.io/Interspeech2026-Audio-Encoder-Challenge/
Join us at Interspeech 2026 in Sydney and benchmark your audio model on the global stage.

The Interspeech 2026 Audio Encoder Capability Challenge, hosted by Xiaomi, University of Surrey, Tsinghua University and Dataocean AI, evaluates pre-trained audio encoders as front-end modules for LALMs, focusing on their ability to understand and represent audio semantics in complex scenarios.

12/04/2025

✨Day 2 at NeurIPS is wrapped!
Thanks to everyone who joined our spotlight session today:
🎤 “Dolphin – A Large-Scale ASR Model for Eastern Languages.”
Speaker: Xiaofeng Xin, General Manager, DataoceanAI
Great energy and great discussions — thank you for the support!
We’re back tomorrow with more demos and conversations.
Four days to go and plenty more to share.
Meet us at SILVER Pavilion – Booth #6.

12/03/2025

🚀 NeurIPS 2025 | Day 1 Highlights
Dataocean AI is live at SILVER Pavilion – Booth #6.
From real-time demos to deep conversations on multilingual & multimodal data, it’s been amazing to meet so many researchers, builders, and innovators.
Thanks to everyone who stopped by to explore our latest datasets！
See you tomorrow — we’ll have more live presentations on site.

11/25/2025

🚀 Meet DataoceanAI at NeurIPS 2025!
📍 SILVER Pavilion – Booth #6
Explore our latest multilingual & multimodal datasets and live demos.
🎤 Spotlight Talk:
“Dolphin – A Large-Scale ASR Model for Eastern Languages” Dec 3 · 10:30–10:42 · Exhibition Hall A
🤝 Let’s connect and accelerate your AI innovation!

11/03/2025

🌍 Unlock the Power of Multilingual OCR Datasets with Dataocean AI!
From natural scenes to handwritten documents, DataoceanAI provides diverse, high-quality OCR datasets to accelerate model training and expand global application coverage.
📘 Available Datasets:
10 Languages Natural Scene & Document OCR Dataset — 45,000 images
9 Languages OCR Dataset — 2,200 images
Thai Natural Scene OCR Dataset — 14,000 images
Japanese Handwriting OCR Dataset — 3,200 handwritten samples
💡 Explore how DataoceanAI helps you build smarter, more accurate, and more inclusive AI systems.
👉 Contact us via email to learn more!

10/15/2025

GITEX GLOBAL 2025 Day 3 — The excitement continues! 🚀
💬 Visit us at Booth H14-A60!
We connected with global clients, partners, and industry experts to explore how high-quality data drives the future of intelligent applications.
Our ASR, TTS, and Multimodal Datasets attracted strong interest from visitors eager to advance AI innovation through better data. 👍

Looking forward to more meaningful connections in the coming days. 🙌

09/11/2025

💡 What if your AI could interrupt you naturally—just like a real conversation?
🔹 Train with Dataocean AI’s 9,000-Hour Chinese Full-Duplex Corpus — powering the next generation of real-time, interruptible AI.
✅ 10,000 speakers across diverse scenarios
✅ Rich annotations: interruptions, overlaps, laughter, feedback cues
✅ Diverse scenarios: daily conversations, business meetings, AI assistants, new energy scenarios, and more
✅ High transcription accuracy: up to 97%
🚀If you want your models to reach GPT Realtime–level fluency, this dataset is your starting point.
👉 Explore the full story here:

Currently, most speech training datasets consist of continuous recordings with complete conversational turns, lacking the naturally occurring, hard-to-model

08/28/2025

🔥 Level Up Your Mandarin ASR!
🔊 9,000 Hours Chinese Mandarin Full Duplex Speech Recognition Corpus (Mobile & Desktop) — our most popular dataset for building smarter, more natural conversational AI.
🚀These datasets are widely adopted for ASR, dialogue systems, and enterprise AI training, helping teams build more natural and reliable conversational experiences.
👉 Want to learn more? Let’s connect! Email: [email protected]