Oriel — AI & technology worth noticing, every day.

gradium/phonon-31m-en

AI model · https://huggingface.co/gradium/phonon-31m-en

First seen by Oriel on .

AI summary

The gradium/phonon-31m-en model provides streaming, 24 kHz English text-to-speech designed for private, offline use on browsers, phones, and laptop CPUs.

  • A 31M-parameter flow-matching transformer predicts audio latents frame by frame, while the Mimi codec decoder produces mono 24 kHz speech.
  • Its 64.5 MB model.q8.gguf file is 8-bit quantized and runs through command-line, Python, browser, and Transformers integrations.
  • In English evaluations, phonon-31m-en achieved 2098 Elo, 1.5% Hard set LLM WER, and 0.51% Seed-TTS LLM WER.
  • The model supports approximately 100 catalog voices, streaming generation, local OpenAI-compatible serving, and applications including screen readers, games, and Home Assistant.
  • It only speaks English, requires matching language normalization, and splits very long inputs so prosody does not span the entire text.

#text-to-speech #audio streaming #edge ai #speech synthesis

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: 120, latest observation ; +11 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: 53,480, latest observation ; +14,010 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Modelsgradium/phonon-31m-enhttps://huggingface.co/gradium/phonon-31m-en Read discussion