On November 17, 2026, Kotoba Technologies Japan announced that Kotoba Technologies has agreed to be acquired by voice AI company ElevenLabs. Kotoba's 15 employees will join ElevenLabs and work to expand the adoption of voice agents and voice interfaces in Japan. It is ElevenLabs' first acquisition of a Japanese company.

Kotoba's voice model, Koto, is already used for Japanese-language phone support. Through the acquisition, the companies plan to combine Kotoba's expertise in East Asian languages with ElevenLabs' computing resources and business infrastructure, accelerating research and development while strengthening sales and deployment support in Japan. The deal is expected to close in the second half of the fourth quarter of 2026, subject to conditions including regulatory approval.

AD

The technology behind Koto's Japanese listening and speaking

For AI to handle a customer call, it must accurately understand what the other person says and respond with appropriate speech. Speech-to-text (STT) converts spoken words into text, while text-to-speech (TTS) generates speech from text. Koto is used for both processes in Japanese phone support.

According to Kotoba's announcement on August 10, Salesforce's Agentforce Voice launched in Japan on August 7 and uses Koto's Japanese speech recognition and synthesis technology. Agentforce Voice is a feature that supports phone conversations between AI agents and customers. Koto handles listening to the customer's Japanese and converting the AI agent's replies into speech.

According to Salesforce's developer documentation, Agentforce Voice lets users select the speech recognition and speech synthesis models separately. For Japanese, "Kotoba STT" is available for speech recognition and "Kotoba TTS" for speech synthesis, and changing one model does not affect the other's settings. This setup lets companies evaluate listening and speaking separately and choose the models best suited to their needs.

Kotoba also says Koto includes technology for simultaneous interpretation that converts speech directly into other speech. In addition to the speech recognition and synthesis used in Japanese phone support, real-time speech translation falls within the company's technical scope.

Existing phone support services and post-acquisition plans

After the acquisition closes, the two companies plan to combine the East Asian language expertise Kotoba has built up with ElevenLabs' research and development capabilities and computing resources. Koto's Japanese speech recognition and synthesis technology was available before the acquisition was agreed, so the deal is meant to build on existing technology by expanding R&D and deployment support in Japan.

The table below summarizes the current service status and post-acquisition plans, based on the companies' announcements and Salesforce's developer documentation.

Status Date Details Primary source
Existing service Launched August 7, 2026; announced August 10 Agentforce Voice launched in Japan, using Koto's Japanese speech recognition and synthesis technology for phone support Kotoba announcement
Current setup Confirmed November 17, 2026 Speech recognition and synthesis models can be selected independently. Kotoba STT and Kotoba TTS are listed for Japanese Salesforce developer documentation
Post-acquisition plan Posted November 16, 2026 Integrating Kotoba's East Asian language expertise with ElevenLabs' R&D capabilities, computing resources and global business infrastructure Kotoba announcement
Japan business plan Announced November 17, 2026 Over the next few months, set up a team to lead domestic sales and continue investing in infrastructure in Japan Kotoba Technologies Japan announcement

After the acquisition, the Kotoba team will continue to handle responding to Japan's language, culture and customer needs. The plan is to apply Kotoba's expertise to product development and deployment support while drawing on ElevenLabs' R&D foundation. Strengthening the domestic sales structure is also intended to bring technology already used in phone support to more companies.

AD

How stronger R&D could improve Japanese phone support

In judging the effects of expanded R&D and deployment support, what matters is how naturally AI can converse on real phone calls. Kotoba cites challenges specific to Japanese voice AI, including dropped subjects, honorifics that change depending on the relationship with the other person, and the timing of backchannel responses and pauses in conversation. The company explains that it is not enough to simply recognize words and generate speech; the AI must understand the context of the conversation and respond at the right moment.

Companies deploying the technology can check separately whether the AI correctly recognizes customer speech and whether its replies sound natural. In its developer documentation, Salesforce also describes a testing procedure in which users listen to synthesized voices in a browser and then call the AI agent through the voice channel customers will actually use. Because audio samples and real phone calls sound different, verification in the actual usage environment is necessary.

The announcement does not disclose the specific model configuration after the acquisition or measured performance improvements. However, because Agentforce Voice currently lets users select speech recognition and synthesis models independently, companies can evaluate the performance of each through actual phone support. If improvements to Koto using ElevenLabs' R&D resources progress, the results could be checked in areas such as recognition accuracy, naturalness of speech and response timing.