# Dataocean AI

> Dataocean AI, formerly Speechocean, supplies licensed datasets and custom data collection, annotation, and evaluation services for speech, language, vision, autonomous driving, and generative AI. Founded in 2005, it combines multilingual collection resources with its DOTS production platform and collaborates with Tsinghua University on the open-source Dolphin speech-recognition models.

- **Founded:** 2005
- **Headquarters:** Spokane, United States (US headquarters; founding and listed legal entity based in Beijing, China)
- **Founders:** He Lin (贺琳) (Founder; founding shareholder and historical chairwoman), Yu Ge (俞舸) (Co-founding shareholder, as recorded in the IPO prospectus)
- **Products:** Off-the-shelf datasets, Custom data collection and annotation, DOTS platform, Generative AI data services, Model training and evaluation
- **Notable:** Founded in 2005; expanded from speech into vision and NLP in 2009, autonomous driving in 2017, and generative AI in 2022., Founding Speechocean legal entity listed on the Shanghai STAR Market in August 2021, ticker 688787., Co-developed openly available Dolphin ASR models with Tsinghua University.

## Products & services

- **Off-the-shelf datasets** — Licensable speech, image, text, and multimodal datasets. Homepage headline advertises more than 1,800; other company pages carry older, lower totals.
- **Custom data collection and annotation** — Recruitment, corpus design, recording, transcription, image and video annotation, point-cloud annotation, quality review, and delivery.
- **DOTS platform** — Configurable data workflows, automated pre-labeling, assisted annotation and quality inspection; the company advertises 200+ task types and 400+ proprietary algorithms. The 2017 timeline milestone refers to its one-stop processing platform, without establishing the DOTS brand launch date.
- **Generative AI data services** — Expert annotation and multimodal data for supervised fine-tuning, preference learning, RLHF, and DPO. Company timeline dates entry to 2022.
- **Model training and evaluation** — Services supporting development and assessment of speech, vision, and language models.
- **Dolphin** — Open-source ASR collaboration with Tsinghua University covering 40 Eastern languages and 22 Chinese dialects, trained on more than 210,000 hours of proprietary and public audio. Code and weights released under Apache 2.0.
- **Dolphin-CN-Dialect** — Streaming and non-streaming speech recognition focused on Mandarin and Chinese dialects, with revised tokenization, sampling, and hotword support.

## Achievements

- Founded in 2005; expanded from speech into vision and NLP in 2009, autonomous driving in 2017, and generative AI in 2022.
- Founding Speechocean legal entity listed on the Shanghai STAR Market in August 2021, ticker 688787.
- Co-developed openly available Dolphin ASR models with Tsinghua University.
- Reports delivery of approximately 2,000 hours of natural duplex conversation within five months for an unnamed technology customer.
- Reports more than 50,000 hours per month of usable-data supply capacity in an unnamed embodied-AI program.

## Latest updates

- **2026-09** — Published embodied-AI, intelligent-cockpit, 3D Gaussian Splatting, and natural duplex-conversation production case studies.
- **2026-07** — Published a case study on emotional TTS data delivered across 14 languages or accents.
- **2026-05** — Released Dolphin-CN-Dialect variants and a paper describing dialect-balanced training and streaming inference.

## Links

- Website: https://dataoceanai.com/
- LinkedIn: https://www.linkedin.com/company/dataoceanai/
- Twitter/X: https://twitter.com/DataOceanAI
- GitHub: https://github.com/DataoceanAI
- Facebook: https://www.facebook.com/DataOceanAILtd

---

Profile page: https://yespress.io/dataocean-ai
Published by YesPress — https://yespress.io
Last updated: 2026-10-23