text-to-speech returns a hosted audio URL and the settings used to make it. The public schema now lists content_type, file_size_bytes, voice, model, speed, input_chars, and source alongside the URL and format.

Keep those values with the script revision. A later narration can reuse the same voice, model, speed, and container while changing only the text.

source reports morpheus or venice, the provider that produced the file. The route still accepts mp3, wav, opus, aac, and flac.