Flat mirror of the italian int8 ONNX bundle for Pocket TTS (Kyutai CALM zero-shot voice-cloning TTS), repackaged for the Sokuji Local Native provider's standard model-repo download path. Source bundle: the KevinAHM/pocket-tts-web Space (onnx/italian/), whose export scripts and licensing apply.

Contents: five int8 ONNX graphs (mimi encoder/decoder, flow-LM main/flow, text conditioner), tokenizer.model, bundle.json, bos_before_voice.npy, voices.bin (eight predefined voices as flow-LM KV-cache prefixes), and voices/manifest.json (the voice list Sokuji's UI reads).

Model weights: CC-BY-4.0 (attribution: Kyutai's Pocket TTS; ONNX export by KevinAHM). Do not use these files outside the terms of that license.