Edge0 releases Audio8-TTS-Realistic, a 0.6B open-weight voice cloning model
Edge0 has released Audio8-TTS-Realistic, an open-weight text-to-speech model designed for zero-shot voice cloning. The 0.6B-parameter model supports 23 languages, adding 12 to its predecessor, and is available under the Apache 2.0 license. It requires a reference audio clip and transcript to clone a voice, with weights and code shipped together for easy deployment.
Key points
- Edge0 released Audio8-TTS-Realistic, a 0.6B open-weight TTS model supporting 23 languages under Apache 2.0.
- The model reports a 4.86% average error on CV3, beating Qwen3-TTS-1.7B but trailing 4B-class models.
- Deployment requires a CUDA GPU or Apple M2 CPU, with peak memory usage of 1.1 to 1.2 GiB in INT4.
The architecture uses a dual-autoregressive design inspired by Fish Audio S2 Pro, built on a Qwen2.5 0.5B base. Edge0 reports that the model achieves a 4.86% average error rate on the CV3 benchmark across 9 languages, outperforming larger open models like Qwen3-TTS-1.7B and VoxCPM2. However, performance on lower-resource languages remains weak, with a 48.57% word error rate on Thai. The company notes that while the model is competitive, 4B-class models like Higgs TTS 3 still lead on certain benchmarks.
For developers, the model is optimized for efficiency, running on a single CUDA GPU with peak memory usage of 1.1 to 1.2 GiB on an Apple M2 CPU in INT4 format. Edge0 emphasizes that the training process includes a learned realism critic to improve naturalness beyond standard word error rate metrics. A free playground is available on Hugging Face ZeroGPU for testing.
Edge0 Releases Audio8-TTS-Realistic, a 0.6B Voice Cloning TTS Model
MarkTechPost · 30 October 2026
Loading the full article…
This text was published by MarkTechPost and written by Asif Razzaq. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Google launches Gemini 4 Carbon for cybersecurity defenders via Fairwind at $2/$10 per million tokens · 12 src
- Google’s Gemini 4 Carbon ranks #1 on Artificial Analysis Intelligence Index v5 with score of 59 · 4 src
- OpenAI launches GPT-6.1 Terra for faster, better-performing AI agents at half the price · 19 src
- Baseten releases 2B-parameter decision model joey-1 at $0.01 per million tokens · 2 src
- Artificial Analysis ranks Claude Opus 5.5 first in updated Intelligence Index v5 · 2 src
Comments
via GitHub Discussions