Silencio-2

Use it to generate expressive English speech, including cloned voices, for applications that need locally run or streamed audio output.

Availability

Weights and inference package are available on Hugging Face

Pricing model

Not established

Deployment

Runs on CPU; described as on-device

Privacy

Not established

Licence

Weights: CC-BY-4.0; code in silencio2/core: MIT

• Generates English speech at 24 kHz from text. ^[1]

  Silencio-2 turns English text into 24 kHz speech in any voice, from a few seconds of reference audio.

• Supports emotion and sound tags in text and voice cloning from a short clean audio clip. ^[1]
• The model weights and inference package are available on Hugging Face.
• The model runs about 20x faster than real time on four CPU threads of an Apple M4. ^[1]

  It runs about 20x faster than real time on four CPU threads of an Apple M4, and all weights fit in one 112 MB file.

• The model supports streaming speech output. ^[1]
• The weights are released under CC-BY-4.0 and the code in silencio2/core is MIT. ^[1]

  The weights are released under [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/), the license of Pocket TTS.