Is Azure Speech-to-Text API Charging Based on Audio Length or Duration When No Audio Data is Sent During Pauses?

I am using WebSocket to capture audio input from the user’s microphone.

This audio data is then sent for continuous recognition (speech-to-text).

My application has the functionality to pause the microphone. When paused, since Azure Speech-to-Text doesn’t provide a pause function, I stop the recognition. After the user clicks the continue button, I restart the recognition.

However, restarting the recognition takes about 3-5 seconds, causing a delay for the user.

I am considering not stopping the recognition API. Instead, I might not send any audio data during the pause. In this case, would the API charge me based on the length of audio processed, or the duration from start to stop?