Edge0 open-sources Audio8-ASR-Infinite streaming speech recognition model
Constant memory and latency over unlimited audio length is the key differentiator, since most ASR systems degrade or overflow on hour-long streams. Developers building live captioning, meeting transcription or real-time voice agents can run it locally without per-session length limits. Bilingual Chinese-English coverage makes it a practical alternative to cloud ASR APIs for regional streaming and voice products.
Edge0 has open-sourced Audio8-ASR-Infinite, a native streaming speech recognition model supporting Chinese and English.
It achieves transcription latency of a few hundred milliseconds and runs 7x24 hours nonstop.A circular sliding cache keeps compute latency and GPU memory usage constant, addressing drift and memory overflow on long audio.
The model has built-in semantic activity detection that distinguishes real pauses, stutters and end-of-speech, targeting live-streaming transcription and low-latency voice interaction.
Edge0开源Audio8-ASR-Infinite：超低延迟实现中英文7x24小时不间断语音转写
这是一款面向实时场景的原生流式语音识别模型，支持中英双语，转写延迟低至几百毫秒。它借助循环滑动缓存机制，使计算延迟和显存占用始终保持恒定，解决了长音频转写容易漂移和内存溢出的痛点。模型还内置语义活动检测，能精准识别真实停顿、结巴与说话结束，特别适合长时间直播听写与高响应实时语音交互等场景。
模型：https://t.co/hUKH7yMpYE
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Audio8-ASR-Infinite, Edge0, speech recognition, open-source models.
Source attribution: curated from Gorden Sun (@Gorden_Sun). Cross-source corroboration: 2 independent sources have covered this development (including AI Will), confirming industry-wide signal strength.
- AI Will09-23 15:48Source
