Emerging Interfaces
Audio8 Transcribes Forever at Constant Memory, and Knows a Thinking Pause From a Finished Sentence
Edge0's Audio8-ASR-Infinite is a 4B streaming speech recognition model whose rolling KV cache keeps memory and latency constant even in 24/7 operation. It pairs a Voxtral Realtime audio tower with a Qwen2.5-3B-Instruct decoder and adds semantic VAD that distinguishes thinking pauses, stuttering and genuine end of turn. Apache 2.0.