Points to Note

Result Consistency

With AmiVoice API, sending the same audio may not always yield the same result. This is because the speech recognition engine is updated daily, and the calculation load is slightly adjusted based on the load status of the speech recognition server.

Originally, speech recognition outputs results probabilistically using statistical information, so applications that need to accept various inputs cannot expect too much consistency. Please design your application so that it does not heavily depend on the text representation of speech recognition results. If you want to implement voice commands with limited vocabulary, consider using the rule grammar of AmiVoice API Private.

API Response Latency

AmiVoice API automatically scales the underlying system according to the load. Therefore, we cannot guarantee that the response time will be constant. If you want to reduce the latency of API responses, please consider using the thread operation mode on a dedicated server with AmiVoice API Private.