Search
On April 24, StepFun introduced its latest automatic speech recognition model, the StepAudio 2.5 ASR. This innovative model takes the lead in incorporating large language model inference acceleration technology into the realm of speech recognition. As a result, it achieves a remarkable 400% surge in inference speed, a substantial 60% reduction in latency, and an impressive 80% decrease in inference costs. It has the capability to transcribe audio recordings of up to 30 minutes in length in one go, showcasing significant enhancements in both speed and accuracy. The model is mainly tailored for scenarios including meeting transcription, voice interaction, input methods, media content processing, and long-audio recognition. Presently, the model is fully accessible on the StepFun open platform and Step Plan.
-
C114 Communication Network -
Communication Home
