Oriel — AI & technology worth noticing, every day.

Edge0/Audio8-TTS-Realistic-Air

AI model · https://huggingface.co/Edge0/Audio8-TTS-Realistic-Air

First seen by Oriel on .

AI summary

Audio8-TTS-Realistic-Air is a compact 0.6B text-to-speech model supporting zero-shot voice cloning across 23 languages, with improved realism over Audio8-TTS-Preview-0.6b.

  • Its DualAR architecture has a 24-layer Slow AR transformer, 4-layer Fast AR transformer, and a bundled 44.1 kHz neural audio codec.
  • It generates speech with or without reference audio; voice cloning requires a clean recording and an exact transcript of its spoken content.
  • An SGLang Omni adapter provides paged attention, dynamic batching, fixed Fast AR KV caching, and an OpenAI-compatible /v1/audio/speech API.
  • Realistic Air recorded 1.25% Seed-TTS error, 4.86% CV3 nine-language error, and 5.89% MiniMax 24-language error, though evaluation pipelines differ across projects.
  • Quality varies substantially by language, while noisy references, transcript mismatches, and finite context can cause pronunciation errors, repetitions, omissions, reduced similarity, or early stopping.

#text-to-speech #voice cloning #transformers #sglang omni

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Models: 407, latest observation ; +177 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Models: 17,750, latest observation ; +10,491 over observed period since

Aliases

SourceExternal IDURL
Hugging Face ModelsEdge0/Audio8-TTS-Realistic-Airhttps://huggingface.co/Edge0/Audio8-TTS-Realistic-Air Read discussion