Edge0 Releases Audio8-TTS-Realistic Voice Cloning Model
Edge0 launches Audio8-TTS-Realistic, a 0.6B parameter open-weight zero-shot voice cloning TTS model supporting 23 languages under Apache 2.0.
Stock photo for illustration only, not from the actual event
- Edge0 unveils Audio8-TTS-Realistic, a 0.6 billion parameter voice cloning model
- Supports 23 languages with zero-shot voice cloning capabilities
- Released publicly under the permissive Apache 2.0 open-weight license
- Features a final GRPO training stage driven by a learned realism critic
The artificial intelligence voice synthesis landscape expands further as Edge0 officially introduces Audio8-TTS-Realistic. This open-weight Text-to-Speech (TTS) model features 0.6 billion parameters, specialized in zero-shot voice cloning, and is made available under the Apache 2.0 license for the broader research and development community.
A major highlight of this release is its multilingual capability, supporting a total of 23 languages. This allows developers to integrate natural-sounding synthesized speech into global applications and platforms without requiring extensive custom voice datasets for every single target language.
Stock photo for illustration only, not from the actual event
Furthermore, the development team incorporated a final Group Relative Policy Optimization (GRPO) training stage. This stage is specifically driven by a learned realism critic, ensuring the generated speech output achieves higher naturalness and significantly minimizes robotic artifacts.
The integration of reinforcement learning techniques like GRPO into audio synthesis highlights a shifting industry trend. Instead of merely scaling up model sizes, modern AI development increasingly focuses on rigorous post-training refinement to bridge the gap between machine-generated audio and human speech.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment