Pocket TTS Loom Packages 100M Model for Edge Devices

This is a useful deployment-focused export rather than a new TTS model: Loom makes Pocket TTS easier to embed in local and edge applications, though the weights themselves remain unchanged.

• –Bundled graph topology, tokenizer, and driver reduce integration work for developers.
• –CPU-friendly inference makes it suitable for offline assistants, accessibility tools, and embedded apps.
• –Kyutai reports roughly 200ms to first audio and 6× real-time on an M4 CPU, but Loom-specific phone benchmarks are not documented.
• –Only Alba is built in; additional voices are separate files with varying attribution and commercial-use requirements.
• –The project is English-only, and “quantized” overstates the model card’s novelty because it describes the weights as unmodified.