Microsoft's Azure AI Speech needs just seconds of audio to spit out a convincing deepfake

Microsoft upgraded Azure AI Speech’s personal voice feature to the zero-shot “DragonV2.1Neural” model, delivering more natural, expressive voice cloning across 100+ languages from only seconds of audio, and added measures like audio watermarks and strict usage policies (consent, disclosure, no impersonation). While the capability enables applications such as customized chatbot voices and multilingual dubbing, the report underscores growing risks of audio deepfakes, citing Consumer Reports’ concerns over insufficient safeguards and an FBI warning about scammers using synthetic voices in fraud campaigns.