Oriel — AI & technology worth noticing, every day.

n4ze3m/tiny-voca

AI model · https://huggingface.co/n4ze3m/tiny-voca

First seen by Oriel on .

AI summary

tiny-voca is a 4.5-million-parameter English text-to-speech model with two fixed voices, designed for fast CPU, browser, and phone inference.

  • It produces 24 kHz mono audio for American English voices F2 and M2, using text phonemized by dizione.
  • The model runs about 70 times faster than real time on four CPU threads and includes PyTorch and ONNX exports.
  • Its transformer-based architecture predicts durations, prosody, and source-filter waveform parameters, avoiding a neural vocoder; the waveform head has 25,862 parameters.
  • Training used 16,018 synthetic clips totaling 17.6 hours, with no real-speaker recordings or pretrained weights, and cost $9.56 in GPU time.
  • It supports utterances up to 512 tokens and 32 seconds, but has no streaming, voice cloning, prosody sampling, or reliable handling of unusual words and foreign names.

#text-to-speech #onnx #speech synthesis #edge inference

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: 200, latest observation ; +80 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: 860, latest observation ; 0 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Modelsn4ze3m/tiny-vocahttps://huggingface.co/n4ze3m/tiny-voca Read discussion