[Official FAQ] Corrections and follow-up questions welcome.

Whisper-class models turn audio into text; they are not chat models. A common pipeline is mic → ASR → LLM → optional TTS.

Tips:

• faster-whisper or whisper.cpp are popular efficient options; larger models are more accurate but slower.
• Noise, accents, and jargon still cause errors—verify critical transcripts.
• If VRAM is tight, run ASR on CPU or unload it before loading the LLM.

Latency is hardware- and audio-dependent; treat any timing claims as rough order-of-magnitude only.