The hash is computed ONLY using the text that is spoken. At a minimum, the hash needs to be the text to be rendered as speech AND the text-to-speech engine used. Otherwise, the cache may be misused when changing from neural to standard (or vice-versa).
The hash is computed ONLY using the text that is spoken. At a minimum, the hash needs to be the text to be rendered as speech AND the text-to-speech engine used. Otherwise, the cache may be misused when changing from neural to standard (or vice-versa).