Finastro voice: Pocket TTS spanish_24l (ONNX int8)

ONNX files that the Finastro app downloads to speak on the device with the voices of its assistants.

Origin and credits

  • Model: Pocket TTS by Kyutai, languages/spanish_24l at revision 2dd944b099d06bb9edbd221fef138685711d9bf0 (Spanish retrained on 2026-09-24). License CC BY 4.0.
  • ONNX graphs: the export by KevinAHM (pocket-tts-onnx, revision 58a6d00cf13d239b6748cb0769f35c580a8f606c), CC BY 4.0.

Changes

KevinAHM's export is of the earlier Spanish checkpoint. Here the same graphs carry the weights of the current checkpoint (2dd944b): each weight was replaced by its counterpart in the new one, including the transposed copies the graph keeps. The result was checked against the official pocket_tts library. The flow_lm_main, flow_lm_flow and mimi_decoder graphs are quantized to int8 with ONNX Runtime dynamic quantization (MatMul only), as in the original export.

File Content
text_conditioner.onnx Text embeddings (fp32)
flow_lm_main_int8.onnx 24-layer transformer with state (int8)
flow_lm_flow_int8.onnx Flow network, one step (int8)
mimi_decoder_int8.onnx Mimi audio decoder with state (int8)

The matching tokenizer is the tokenizer.model from spanish_24l at revision 2dd944b of kyutai/pocket-tts.

Prohibited use

These are the use terms Kyutai sets for Pocket TTS, and they apply equally to these files:

Use of our model must comply with all applicable laws and regulations and must not result in, involve, or facilitate any illegal, harmful, deceptive, fraudulent, or unauthorized activity. Prohibited uses include, without limitation, voice impersonation or cloning without explicit and lawful consent; misinformation, disinformation, or deception (including fake news, fraudulent calls, or presenting generated content as genuine recordings of real people or events); and the generation of unlawful, harmful, libelous, abusive, harassing, discriminatory, hateful, or privacy-invasive content. We disclaim all liability for any non-compliant use.

License

CC BY 4.0. Credit Kyutai (Pocket TTS) and KevinAHM (ONNX export) when redistributing.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for digitalastrolite/finastro-voz

Quantized
(67)
this model