Articles by tag: ASR

llama.cpp: ASR in Gemma 4 and Qwen3

llama.cpp integrated speech recognition Gemma 4 (E4B/E2B) and Qwen3. Launch on GPU, prompts, parameters for stability. Instructions for middle/senior dev.

NPU on MCU for speech: STM32N6 and acoustics

Acoustic speech recognition model on STM32N6 with NPU: PER 5.3%, 0.2 W, real time. Comparison, metrics, application for developers. Study the implementation.