Kyutai ships Pocket TTS Small: CPU voice cloning from just 1.9M parameters
BerryBlush_Rogers691 · reddit · 2026-10-21
Kyutai released three Pocket TTS models on Oct 20 — 1.9M, 6.9M and 18.2M parameters — English-only, voice-cloning, and designed to run on CPU, with a demo to switch between sizes. The practical win: keep speech on CPU and leave all VRAM to your LLM. The author suggests starting with 6.9M and benchmarking with a few minutes of names, numbers, and awkward punctuation. Note: the parameter counts include the audio decoder; the reference-voice encoder is separate and runs only once. Code and setup instructions are on GitHub.
More from Infra
- GLM-5.3-Flash NVFP4 benchmarked: 2x H200 matches 4x RTX Pro 6000 on throughput — TheZachMueller · 2026-10-21
- DeepSeek V4.1 Flash inference teardown: Engram cuts 30% params, WideEP spreads 552B over 64 GPUs — teortaxesTex · 2026-10-21
- llm-d 1.0 ships: multi-tenant QoS, zero-downtime updates, prefix-cache-aware routing — _ScottCondron · 2026-10-21
- DeepInfer math: 1GW of GB200 could generate $100B+ revenue at DeepSeek's current prices — bookwormengr · 2026-10-21
- FastVideo releases FastH3 Omni weight collection with FP8 and NVFP4 ComfyUI builds — multimodalart · 2026-10-21
- Voice models should run on-device: Phonon open-source release sparks debate — timshi_ai · 2026-10-21