Open real-time speech translation with voice cloning puts interpreting and offshore call centers on notice

Real-time voice translation, including in the speaker's own voice, is already a product category. Sanas and Krisp sell it to contact centers, Transcom has deployed it across its BPO operations, and Microsoft Teams and Google Meet include it, with Teams already covering Mandarin. The base capability is no longer a moat. One opening is a sourcing-call tool for Chinese suppliers. Teams and Meet only work inside their own meetings, while suppliers mostly talk over WeChat and phone. Such a tool could win on Chinese and accent handling, noisy factory floors, terms like price, MOQ (minimum order quantity), Incoterms and specs, and records that hold up in a dispute: bilingual transcripts, extracted commitments and a summary for the purchase order. No open model translates Chinese in real time while keeping the voice, so a newcomer would have to build or fine-tune that part. For BPOs, Krisp and Sanas sell the translation layer, but I found nobody selling the operations around it: scheduling one pool of agents across languages, sampling translated calls for quality, scoring accuracy by language and keeping audit trails. Running fully on-premises could appeal to buyers who can't send call audio to the cloud, but that is unproven. The main risk is that Krisp, Sanas or the call platforms add these features themselves, so a newcomer has to win on Chinese-language quality and workflow rather than on translation alone.

• Sanas Real-Time Translation

  Launched in June 2025 for contact centers. Translates speech directly to speech and says it keeps the speaker's identity and emotion. Started with English-Spanish and English-French. Targets BPO agents directly.

• Krisp AI Live Interpreter and Voice Translation SDK

  Voice-to-voice translation used in live contact centers since 2025. An SDK released in February 2026 covers any pair among 60+ languages. Sold with accent conversion and noise cancellation in Krisp's contact-center platform.

• Transcom with Avaya and Sabio

  A large BPO already runs real-time speech translation in 100+ languages so it can hire agents for knowledge rather than languages. This is essentially the language-pooling idea, already deployed.

• DeepL Voice for contact centers

  Real-time transcription and translation for contact-center agents. Its voice-to-voice version was described as upcoming.

• Microsoft Teams Interpreter agent

  Real-time speech translation in Teams meetings that can imitate the speaker's voice. Includes Mandarin Chinese. Requires a Microsoft 365 Copilot license and includes 20 hours per user per month.

• Google Meet speech translation

  Gemini-powered live translation in the speaker's own voice, starting with English and Spanish and adding more languages.

• Kyutai Hibiki / Hibiki-Zero

  Open-weight real-time speech translation with voice transfer that can run on a device. Only translates into English, from French, Spanish, Portuguese and German. No Chinese.

• Meta Seamless (SeamlessStreaming / SeamlessExpressive)

  Open research models with about 2 seconds of delay across roughly 100 languages that preserve vocal style. A base someone could build a local product on.

• Mentalese

  Open-source proof of concept for local two-way voice translation with voice cloning that sits between the user and apps like Discord or Meet. Not production-grade.

• Soniqo Speech Studio

  On-device speech tools with local speech-to-speech, streaming transcription and zero-shot voice cloning. The Mac app is Apache 2.0 licensed.

• JotMe

  $9/month live call translator for Zoom, Meet and Teams, marketed for Chinese-English. A general tool, not aimed at sourcing.

• Soniox speech translation API

  Streaming Chinese-to-English speech translation API that developers could build a sourcing tool on.

Checked 2026-10-26