memory-modelling

Voice Memory introduces an inference-only scheme for agentic speech recognition where a frozen corrector uses a per-domain memory file to decide per utterance whether to act or abstain, reducing word error rate across multiple domains without weight updates.