Edge0 releases Audio8-TTS-Realistic, a 0.6B open-weight voice cloning TTS model under Apache 2.0

Edge0 has released Audio8-TTS-Realistic, a 0.6B-parameter open-weight zero-shot voice cloning text-to-speech (TTS) model licensed under Apache 2.0. MarkTechPost reported the release on October 29, 2026.

Key specifications

• Model: Audio8-TTS-Realistic
• Developer: Edge0
• Parameters: 0.6B
• License: Apache 2.0 (open weights)
• Capability: zero-shot voice cloning
• Languages: 23
• Training: includes a final GRPO stage driven by a learned realism critic

Context window, pricing, benchmark scores and training cutoff date are not yet disclosed.

Training approach

The model adds a final GRPO (Group Relative Policy Optimization) stage to its training. That stage is driven by a learned realism critic, which means a trained model supplies the signal used to optimize how realistic the generated speech sounds.

What this means

Audio8-TTS-Realistic combines a small 0.6B parameter count with zero-shot voice cloning across 23 languages, released as open weights under Apache 2.0. The final GRPO stage with a learned realism critic shows that Edge0 targets realism directly with a reinforcement-learning step at the end of training. Builders evaluating the model will need independent listening tests and benchmarks, since no evaluation figures have been published alongside this announcement.