Streaming Recognition
A speech recognition approach in which audio is continuously fed to the model in chunks and recognition results are returned incrementally. It requires low latency and supports rolling correction of partial results.
Timeline (last 90 days)
The end-to-end latency of streaming recognition is typically required to be kept within a few hundred milliseconds, and it must support rolling correction of partial results as context evolves.
Dictation requires streaming recognition: audio is continuously fed into the model in windows of tens of milliseconds, with recognition and output happening simultaneously, placing extremely high demands on first-character latency and sentence-segmentation stability.
Streaming recognition requires the inference engine to support stateful computation graphs, dynamic input shapes, and low-latency scheduling strategies.
In recent years, on-device ASR error rates have dropped significantly, and streaming recognition latency has entered the hundred-millisecond range.
Streaming recognition typically maintains a long connection via WebSocket and is suitable for interactive scenarios such as voice input methods and real-time subtitles; asynchronous recognition returns results via polling or webhooks and is better suited to offline scenarios such as recorded notes and meeting minutes.
流式STT和TTS允许系统在处理未完成时就开始并行处理，从而压缩用户感知到的等待时间
All Facts (6)
The end-to-end latency of streaming recognition is typically required to be kept within a few hundred milliseconds, and it must support rolling correction of partial results as context evolves.
50%UnverifiedDictation requires streaming recognition: audio is continuously fed into the model in windows of tens of milliseconds, with recognition and output happening simultaneously, placing extremely high demands on first-character latency and sentence-segmentation stability.
50%UnverifiedStreaming recognition requires the inference engine to support stateful computation graphs, dynamic input shapes, and low-latency scheduling strategies.
50%UnverifiedIn recent years, on-device ASR error rates have dropped significantly, and streaming recognition latency has entered the hundred-millisecond range.
50%UnverifiedStreaming recognition typically maintains a long connection via WebSocket and is suitable for interactive scenarios such as voice input methods and real-time subtitles; asynchronous recognition returns results via polling or webhooks and is better suited to offline scenarios such as recorded notes and meeting minutes.
50%Unverified流式STT和TTS允许系统在处理未完成时就开始并行处理，从而压缩用户感知到的等待时间
50%