Amazon Now Lets You Clone Any Voice With AI in Minutes

Amazon has added a voice-cloning AI tool called Qwen3-TTS to its cloud platform, SageMaker. Here's how it works in plain terms: you hand it a short recording of someone talking — just a few seconds — along with a written transcript of that clip. Then you type in whatever new text you want. The AI reads your new text out loud in that person's voice, copying their pitch, tone, and rhythm. It doesn't need to be trained on hours of audio first, which is what made this kind of thing slow and expensive before.

The tool speaks 10 languages: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian. The clever part is cross-language cloning. Record someone speaking English, and the system can produce their voice speaking Japanese while still sounding like them. Publishers could turn one audiobook into ten versions with the same narrator. Schools could teach in multiple languages using a familiar teacher's voice. Companies could give every customer-service bot the same consistent brand voice.

Amazon's pitch to businesses is control. Instead of paying per-character fees to an outside AI service, companies rent computing power by the hour and their audio never leaves their own account. Amazon handles the servers, monitoring, and automatic scaling, so nobody has to babysit the hardware. That makes this cheap and private enough for companies handling sensitive recordings.

The obvious catch is misuse. This is exactly the technology behind convincing scam calls and fake audio of real people. Because the model is publicly downloadable, anyone with modest technical skills can run it, and there's nothing built in that verifies whether the person being cloned actually agreed. If you get an unexpected call from a family member asking for money, that voice is no longer proof of anything.