: ⁠Feature Request: Distilled 6-layer model for French (french_6l) for local iOS deployment?⁠

Hi Kyutai team, First, thank you for the amazing work on Pocket TTS. I am the developer of Meemaw.app, an iOS application that heavily relies on local, offline voice generation. Our core feature involves dictation exercises ("dictées") read by a grandmother persona. My Current Pipeline & Its Limitations: Currently, my local pipeline uses the native iOS TTS passed through an ONNX Voice Conversion model to achieve the desired grandmother timbre. While it works, the baseline prosody remains robotic. It fails to properly respect punctuation, breathing, and emotion. For dictation exercises, this lack of natural rhythm often makes it difficult for users to understand the context and spell words correctly. Why we need Pocket TTS: Your model perfectly solves the prosody and punctuation issues natively. However, running the ⁠french_24l⁠ model locally on an iPhone (via CoreML or MLX) is currently too heavy and consumes too much RAM for a smooth user experience. The Request: Since the distilled English ⁠en_6l⁠ model is incredibly efficient for edge deployment, I was wondering if a distilled ⁠french_6l⁠ model is currently on your roadmap, or if you might already have internal weights you would consider releasing? Having a 6-layer French model would be an absolute game-changer for local iOS/Apple Silicon developers who need high-quality French TTS without compromising device memory. Thank you for your time and your massive contributions to open science!