Voice cloning goes open source, voice agents go enterprise
The threshold for cloning a voice just moved. Confucius4-TTS needs three seconds of audio and works across 14 languages. Enterprise voice AI got its biggest distribution deal of the year with IBM and ElevenLabs. And a startup building testing tools for voice agents raised $28 million because production deployment is finally the bottleneck.
1. NetEase open-sourced voice cloning across 14 languages
NetEase Youdao dropped Confucius4-TTS on June 23 and it is the kind of release that makes you re-evaluate what “open source voice AI” means. The model clones a voice from three seconds of audio and speaks it across 14 languages with no detectable accent, no reference transcript needed, and no per-language fine-tuning. Single-digit WER across every supported language.
The numbers: 85% voice similarity on a three-second sample, 97% on longer clips, all under MIT license on GitHub. That puts cross-lingual zero-shot voice cloning in the hands of anyone who can run a model. The disclaimer is that the demo clips sound cleaner than what most teams will get on their first try, but the threshold just dropped hard.
2. ElevenLabs partnered with IBM and rolled out SynthID
Two moves from ElevenLabs in the same week. The first was a partnership with IBM that brings ElevenLabs TTS and STT into watsonx Orchestrate, IBM’s agentic AI orchestration platform. The deal gives enterprise customers access to ElevenLabs’ 10,000-voice library with PCI compliance, HIPAA-compatible zero-retention mode, and data residency controls. The second was adopting Google’s SynthID watermarking, baked into every TTS generation for free users and rolling out across all tiers.
The IBM deal is about distribution: enterprises buy IBM’s compliance story, and ElevenLabs gets plugged into procurement cycles that would take years to navigate alone. The SynthID move is about regulatory positioning, because the EU AI Act Article 50 goes live in August and “we watermark by default” is becoming a checkbox requirement.
3. Coval raised $28 million for voice agent testing
Coval announced a $28 million Series A on June 24, led by Norwest with participation from Twilio Ventures and Y Combinator. The company builds testing and monitoring software for voice AI agents: simulations before launch, live performance tracking after, and data labeling in between. Customers include Zoom and Deepgram.
The round is interesting for what it says about the market. Coval’s thesis is that building a voice agent is easy, but getting it right for production across millions of interactions is not. The company claims its platform cuts manual QA work by up to 30x and gets agents to production 10x faster. More than $7 billion went into voice AI in Q1 2026, and Coval is selling the picks and shovels. That is a signal that the industry is past the demo phase and into the deployment phase.
4. The UK admitted its voice cloning laws are not fit for purpose
The BBC reported on June 26 that UK law does not adequately protect people from unauthorised AI voice cloning. Media lawyer Dr Mathilde Pavis told the BBC that legislation is “unfit for purpose” because it was written before the technology existed. The UK relies on a patchwork of copyright, data protection, and trademark law that comes close but does not actually cover a person’s voice as a protected attribute.
The story was about Bristol voice artist Faye Dicker, whose voice was cloned and sold online through Fish Audio without her consent, downloaded more than 900 times. Fish Audio removed the content, but the legal framework had no clear remedy. France, Italy, the Netherlands, and Denmark have personality rights that cover voice as identity. The UK does not. The consultation the government promised will take months. This gap matters, and it matters more as voice cloning gets cheaper and more accurate.
5. Speechify shipped live operations APIs and voice typing
My employer’s week: Speechify Voice Agents got live operations APIs on June 22, exposing transcript streaming, call take-over actions, analytics queries, and saved dashboard views. Two days later, Speechify brought real-time voice typing to every iPhone and Mac user, with agentic workflows coming in the following weeks. The 9to5Mac coverage on June 23 also previewed Speechify Work, a productivity product with agentic features, and confirmed a new generation of the Simba API with multilingual support, emotion control, and zero-shot voice cloning.
I joined on a Monday. By Wednesday the product was in the press for something I had not even seen yet. That pace is the norm here.
FAQ
What is Confucius4-TTS and why does it matter?
Confucius4-TTS by NetEase Youdao is an open-source text-to-speech model that clones a voice from three seconds of audio and speaks it across 14 languages with no detectable accent. It matters because it puts cross-lingual zero-shot voice cloning under an MIT license, the most permissive open-source terms available.
How does ElevenLabs’ IBM partnership change enterprise voice AI?
ElevenLabs integrated its TTS and STT into IBM watsonx Orchestrate, giving enterprise customers access to premium voice quality with PCI and HIPAA compliance through IBM procurement channels. The deal bypasses the long sales cycles that voice AI vendors typically face in enterprise markets.
What is Coval and what does it do for voice agents?
Coval is a testing and monitoring platform for voice AI agents that runs simulations, tracks live performance, and labels training data. Its $28 million Series A from Norwest, Twilio Ventures, and Y Combinator signals that the voice agent industry is moving from prototyping to production deployment.
What did the BBC report about UK voice cloning laws?
The BBC reported that UK law does not adequately protect individuals from unauthorised AI voice cloning. Unlike France, Italy, the Netherlands, and Denmark, the UK lacks personality rights that treat voice as a protected attribute of identity.
What shipped at Speechify in my first week?
Speechify Voice Agents added live operations APIs for transcript streaming and call intervention, real-time voice typing launched on iPhone and Mac, and the Simba API roadmap was previewed with multilingual support and zero-shot voice cloning.