Creates a speech recognition (transcription) task for an audio recording and returns its id immediately with HTTP 201; processing happens in the background. The recording is uploaded as a multipart file in recording. Languages, channel handling, speaker diarization and the task type are set in the request body; remote_identifier and metadata let you correlate the task with your own records. When webhook_url is set the result is posted there once the task completes, otherwise poll the result method with the returned id. A recognition token of the authenticated account may be passed in the X-Recognition-Token header (or as a token query parameter); API keys allowed to use Speech Recognition may omit it. Processing is charged from the account balance. Returns 400 with validation errors when the task cannot be created.