Oriel — AI & technology worth noticing, every day.

Edge0/Audio8-TTS-Realistic-0.6b

AI model · https://huggingface.co/Edge0/Audio8-TTS-Realistic-0.6b

First seen by Oriel on .

AI summary

Audio8-TTS-Realistic-0.6B is a compact text-to-speech model that improves on Preview’s speech realism while supporting zero-shot voice cloning and generation in 23 languages.

  • The main DualAR model has 601,159,424 parameters, uses a separate bundled 44.1 kHz codec, and supports up to 2,048 packed text/audio positions.
  • Its slow autoregressive transformer predicts semantic tokens, while a fast transformer predicts codec codebooks; inference uses custom Transformers code and requires trust_remote_code=True.
  • The model adds 12 languages to Preview’s 11, with quality varying by language and evaluation set.
  • On shared CV3 languages, Realistic records 4.56% error versus Preview’s 4.80%; its Seed-TTS error is 1.25% versus Preview’s 1.23%.
  • SGLang Omni serving supports paged attention, dynamic batching, and an OpenAI-compatible speech API, but throughput and streaming were not benchmarked.

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

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: 229, latest observation ; +127 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: 7,259, latest observation ; +6,446 over observed period since

Aliases

SourceExternal IDURL
Hugging Face ModelsEdge0/Audio8-TTS-Realistic-0.6bhttps://huggingface.co/Edge0/Audio8-TTS-Realistic-0.6b Read discussion