Analysis
Microsoft has launched a new speech-generating AI model that the company says is both cheaper and more accurate than competing voice models from ElevenLabs, Google and xAI, The Information reported. The model is aimed at powering features inside Microsoft's own products -- transcribing conference calls in Teams and patient conversations in Dragon Copilot, its AI product for healthcare providers.
A Moving Target
The launch comes just days after ElevenLabs shipped Eleven v4 and v4 Turbo on September 28, adding expression control, lower latency for voice agents and support for more than 90 languages. Microsoft is entering a market that moved materially in the days before its own launch, which cuts against any claim that its model is simply best today -- it's best against a snapshot that's already out of date.
“- Google: ships voice generation through its Gemini and Cloud Text-to-Speech products, with the scale advantage of bundling into Google Workspace and Android.”
The competitive field spans a few distinct strategies:
- ElevenLabs: the category leader in AI voice generation, reportedly valued near $22 billion with roughly $500 million in annualized revenue; its v4 models now support 90-plus languages and power voice agents across customer service and content creation.
- Google: ships voice generation through its Gemini and Cloud Text-to-Speech products, with the scale advantage of bundling into Google Workspace and Android.
- xAI: has pushed into voice as part of Grok's multimodal features, leaning on its compute advantage from the SpaceX-xAI combination rather than a standalone voice product strategy.
Microsoft's advantage isn't model quality in isolation, it's distribution. Teams has roughly 320 million monthly active users and Dragon Copilot is already embedded in thousands of healthcare systems' clinical workflows; ElevenLabs, despite its revenue and valuation, has to win each enterprise voice contract on its own rather than riding an existing seat count.
For VCs with exposure to voice AI startups, Microsoft's entry is the clearest signal yet that the category's economics will increasingly separate between vertically bundled incumbents and horizontal API players. ElevenLabs shipping v4 just before Microsoft's launch suggests the company already expected this and is racing to lock in developers on API quality before Microsoft's bundled version becomes good enough for most Teams customers.
Microsoft's own claim that its model is cheaper and more accurate is self-reported and untested by any independent benchmark cited in the reporting. Voice model quality comparisons are notoriously benchmark-dependent, and "cheaper" says nothing about whether Microsoft's model matches ElevenLabs' expression control or latency for real-time voice agents, the use case ElevenLabs has built its moat around.
ElevenLabs' $500 million ARR shows enterprise voice AI is already a real revenue category, not a feature, which is exactly why Microsoft is bundling rather than competing on API pricing alone.