audio.cpp update cuts Higgs Audio TTS VRAM use by 48%

audio.cpp landed runtime optimizations: Higgs Audio TTS now runs in about 6GB VRAM, a 48% peak-memory cut; HTDemucs is 2.21x faster on GPU and PocketTTS 2.23x faster on CPU, with several TTS models seeing 6-21% VRAM reductions. The project now supports 110+ audio model families and 190+ variants, and adds an experimental WebUI generation history feature.

• Kokoro gains voice cloning via OpenVoice with ~600MB extra VRAM2026-10-29