Summary
Kotoba is building real-time speech models for East Asian languages — ranging from a real-time translation model that matches professional interpreter quality to ultra-low-latency speech models for agent use cases. Salesforce Ventures is proud to participate in Kotoba’s seed round.
- Founders: Noriyuki Kojima (CEO), Jungo Kasai (CTO)
- Sector: Voice AI / Speech Foundation Models
- Location: San Francisco, California / Tokyo, Japan
The Opportunity
Real-time spoken communication across languages remains one of the hardest unsolved problems in global communication — and nowhere is the demand more acute than in Japan and across East Asia. Despite decades of research, existing solutions tend to be too slow, too inaccurate, or too poorly calibrated for the phonetic complexity of Japanese, Korean, and Chinese. As AI agents proliferate and voice becomes the primary interface of AI, the gap between what global platforms offer and what users need has never been wider — or more valuable to close.
The Solution
Kotoba’s proprietary model, Koto, is purpose-built for real-time speech translation and voice interfaces — with industry-leading performance in Japanese, Korean, and Chinese. Koto works across both end-to-end and traditional architectures, delivering ultra-low latency whether it’s handling the full speech pipeline or slotting in as a best-in-class speech-to-text and text-to-speech layer within an existing LLM stack.
Koto’s simultaneous interpretation app lets people speak naturally in one language and be heard in another, in real time, at broadcast-quality accuracy. On the enterprise side, Koto is available for companies to integrate directly into their own products as voice recognition and speech generation technology — running even on-device, including standard smartphones, without needing a connection to a remote server.
Why We’re Backing Kotoba
Kotoba’s founders are frontier AI researchers who chose to build. That combination is genuinely rare. Noriyuki Kojima (PhD, Cornell) received the Best Paper Award at EMNLP 2022 and co-founded the LLM Fugaku project — Japan’s large-scale language model initiative built on the Fugaku supercomputer. Jungo Kasai (PhD, University of Washington) was an Assistant Professor at Toyota Technological Institute at Chicago (TTIC) and received the Best Paper Award at NAACL 2022, making him one of Japan’s most cited Natural Language Processing (NLP) researchers. Both are Masason Foundation Fellows.
What sets Kotoba apart is a potent combination of research-grade model quality, deep Asian language specialization, and a San Francisco and Japan-based team that possess the development flexibility and responsiveness that enterprise deployment demands. As AI agents become the dominant software paradigm, Kotoba is positioning Koto as the defacto of speech model for next-generation devices and agentic systems in East Asia.
What’s Ahead
We’re proud to participate in Kotoba’s seed round alongside Kindred Ventures. We believe the company’s exceptional founders, technically differentiated models, and momentum across consumer and enterprise make it a category-defining business in the making.
