I separated the audio file delivery and transcription initialization in #speechzap to avoid consuming minutes in case of transcription failure, especially for large files.