[audio.cpp] Recent updates you might have missed: Higgs Audio TTS use 48% less VRAM (< 6GB), HTDemucs 2.2× faster, PocketTTS 2.2× faster on CPU, and WebUI generation history feature

Recent updates to audio.cpp have significantly improved performance for various audio models. Higgs Audio TTS now uses 48% less VRAM, bringing its requirement under 6GB. HTDemucs is 2.21x faster on CUDA and 1.95x faster on Vulkan, while PocketTTS shows a 2.23x speedup on CPU and uses 9% less RAM. Other models like MOSS-TTS and Qwen3-TTS also benefit from VRAM reductions, making these models more practical for local execution.

This report details specific performance improvements, including Higgs Audio TTS using 48% less VRAM and HTDemucs running 2.21x faster, unlike general updates.