Kotoba Technologies, a developer of real-time speech models optimized for East Asian languages, has raised $10 million in seed funding.
The financing was led by Kindred Ventures, with participation from Salesforce Ventures and Sony Innovation Fund. The round brings the company’s total funding to date to $23 million.
The technology
Kotoba’s proprietary model, Koto, is purpose-built for real-time speech applications such as AI
agents, smart hardware devices, and simultaneous speech translation, with industry-leading
performance in Japanese, Korean, and Chinese.
Koto offers flexible deployment options depending on the use case:
- Available as speech-to-speech (S2S) models as well as ultra-low-latency speech-to-text
(ASR)and text-to-speech (TTS) models. - Deployable both in the datacenter and on-device, including on smartphones and
wearables.
Prospects
Kotoba will direct the new funding toward three priorities at the core of its speech AI platform in
East Asia:
- Speech-to-Speech (S2S): Koto has demonstrated sub-2-second latency in
simultaneous translation. Kotoba will invest further in this model family, extending it to
broader use cases such as AI agents and smart devices while continuing to push the
quality of simultaneous translation. - On-Device Rollout: Koto already runs on-device with Kotoba’s enterprise customers in
Asia and US. Kotoba will dedicate more resources to running Koto efficiently on edge
chips and will explore wider distribution channels across automobiles, electronics, and AI
wearables through partnerships. - Agentic Rollout: Kotoba will further improve the usability of the Koto ecosystem for
enterprise customers globally, accelerating their expansion into Asian markets. This
includes model ecosystem development as well as forward-deployment efforts.
Kotoba’s API / SDK Release
Koto is already in production with leading global organizations, including Fortune Global 500
companies and high-growth AI-native startups. In these deployments, the technology powers AI
voice agents, voice interfaces for contact centers, wearable devices, and AI-powered
simultaneous translation.
To broaden developer access, Kotoba has released an alpha version of its API and an easy-to-
use Python SDK. Its S2S simultaneous translation models and ultra-low-latency speech-to-text
and text-to-speech models are now available via API, and on-device models can also be tested
via the API/SDK. Kotoba is committed to further expanding its API/SDK ecosystem. See the API announcement for details.
“Kotoba” Translation App Is Growing Rapidly
Koto is widely used by prosumers and enterprise users across East Asia through Kotoba’s
proprietary app, “Kotoba” (同時通訳). Built on Kotoba’s proprietary model, the app delivers
seamless, flagship-quality simultaneous translation, note-taking, and AI summaries.
It empowers users with real-time multilingual communication across 21 languages (with five
primary target languages) — in business settings as well as entertainment, tourism, and a wide
variety of other scenes. In June, Kotoba shipped a major update introducing 11 new features
[see article] alongside significant UI/UX improvements. The company is also deepening its
enterprise support, with a meeting-agent experience for remote conferences planned for release in July. The Kotoba app has now surpassed 180,000 users, with daily downloads continuing to grow.
Investor Perspectives
Steve Jang, managing partner at Kindred Ventures, said in a statement, “Asia is home to nearly five billion people, and to start, East Asian countries represent 1.6 billion of that
continental population. Roughly half of the world’s knowledge workers speak an Asian language as their first native tongue. The complexities of getting the unique aspects of Asian languages requires a unique training strategy and learning loop approach with a deep understanding of each language and market.”
Jang added, “The Kotoba research team brings extreme focus and depth to developing the world’s fastest and most genuine speech models for both high-controllability pipelines for agents, or incredibly fast and accurate native speech-to-speech models for realtime communication and translation.
He said that on both recognition and synthesis, their Koto family of models – TTS, STT, and Speech-to-Speech – models performed better than existing models developed by American and European research labs.
“We’re thrilled to support Kotoba’s mission to bring state-of-the-art speech models, multimodal agents, voice-centric wearables, physical Al hardware, and the holy grail of realtime translation to the entire world,” Jang said.
Ken Asada, partner and Sho Yamanaka, principal at Salesforce Ventures, said in a statement, “Under a co-founding team that combines exceptional research capabilities with strong business execution, Kotoba Technologies is developing world-class voice AI and steadily advancing its real-world implementation. In addition to their high technical capabilities, we see immense potential in their focus on driving implementation in business environments. We look forward to leveraging Salesforce’s global network and expertise to support the company’s further business growth.”
Austin Noronha, managing director at Sony Ventures-US, said in a statement, “Real-time voice communication remains one of the most technically challenging AI frontiers. Kotoba has demonstrated impressive real-world results in both translation quality and latency, outperforming many existing approaches in speech-to-speech translation. With encouraging
early product-market fit and growing adoption among enterprise customers, Kotoba is building
more than a translation application, it is creating a voice AI infrastructure platform with potential
applications across enterprise, telecom, electronics, and consumer markets.”