Science & Technology

ElevenLabs just got nuked by open source

by Jeff Geerling

📚 Main Topics

Introduction of Quen 3 TTSThe release of Quen 3 TTS by Alibaba Cloud and its implications for voice cloning technology.
Comparison with ElevenLabsDiscussion on the advancements in voice cloning capabilities, particularly in relation to ElevenLabs.
Ease of Voice CloningThe simplicity of cloning voices using open-source tools and the potential risks associated with it.
Personal ExperienceThe speaker shares their personal experience with voice cloning and the unauthorized use of their voice.

✨ Key Takeaways

Voice Cloning TechnologyQuen 3 TTS allows users to clone voices easily by uploading audio and a transcript, making it accessible on various devices, including Raspberry Pi, Mac, and smartphones.
Risks of MisuseDespite safeguards, the technology can be misused, leading to unauthorized voice cloning that can deceive audiences.
Quality of Cloned VoicesWhile cloned voices may lack the full intonation and quirks of a real person, they can still be convincing enough for short phrases, posing a risk for misinformation.
Concerns for Content CreatorsContent creators, like the speaker, express concern over the potential for their voices to be cloned and used without consent, impacting their brand and revenue.

🧠 Lessons

Awareness of TechnologyAs voice cloning technology becomes more accessible, individuals should be aware of the implications and potential for misuse.
Protecting Personal IdentityContent creators need to consider how to protect their voice and identity in an era where cloning is becoming easier.
Ethical ConsiderationsThe rise of AI-generated content raises ethical questions about authenticity and consent in digital media.

Overall, the advancements in voice cloning technology, particularly with tools like Quen 3 TTS, present both exciting possibilities and significant challenges, especially regarding personal identity and content integrity.

qwen tts text to speech model ai llm elevenlabs 11labs labs company open source weights weight tokenizer convert sneakers voice clone cloning steal vocal huggingface hf download upload app ease easy easier

Transcript excerpt

0:00 Quen just released Quen 3 TTS, and ElevenLabs has something to be worried about. I already talked about last year how somebody cloned my voice and used it in tutorial video series, and it sounded pretty good. That was using ElevenLabs last year, and they're a little bit better now. And it's very easy to take someone's audio, like you could just take the audio off of this video, upload it to ElevenLabs, and clone my voice. A lot of these cloud services are supposed to have safeguards against cloning other people's voices, but, you know, those work better and worse.

0:31 But that's besides the point now, because there is an open-source Qwen, which is made by Alibaba Cloud in China. They have released this TTS series of models that you can download and run on almost any kind of system. I can run it on a Raspberry Pi with an external GPU. I can run it on my Mac. I could even run it on my phone if I wanted to. And you just take a recording of someone and you put a transcript in, put those together, and a couple minutes later you can clone someone's voice and make them say anything.

🔒 The full, searchable transcript is available with Pro.

Suggestions