Billing

The Paraformer speech recognition model service transcribes and bills only for the parts of an audio track that are identified as speech. Non-speech content is not billed. The duration of the speech content is usually shorter than the duration of the original audio file. Because an AI model identifies the speech content, this identification may not be perfectly accurate.

For multi-track audio files, only the first track is transcribed and billed by default. If multiple tracks are specified for transcription, each track is billed separately based on the duration of its speech content. The actual billable duration for each request is available in the content_duration field of the response. For more information, see Return parameters.