===== Categories of Thinking Tokens =====

Overview
- The attribute-based thinking tokens are converted into special tokens and fed to the GPT.
- When speech is neutral or uses a standard US/UK accent, Accent and Style are treated as None (i.e., no Accent/Style token is emitted).

Gender:
- male
- female

Emotion Intensity:
- high
- medium
- low

Accent:
- Russian-accented English
- Indian-accented English
- German-accented English
- Spanish-accented English
- British
- Chinese-accented English
- Eastern European
- Colombian
- Italian-accented English
- Hispanic
- Irish
- French-accented English
- Polish-accented English
- Southern US accent
- Middle Eastern-accented English
- European-accented English
- Italian-American
- Scottish
- New Jersey accent
- Arabic-accented English
- Korean-accented English
- Ukrainian-accented English
- Birmingham accent

Style:
- shouting
- hesitant
- whispering
- strained voice
- sobbing while speaking
- muttering
- yelling
- laughing while speaking
- singing
- stuttering
- sing-song tone
- slurred speech
- hoarse voice
- screaming
- monotonic
- breathy voice
- robotic tone
- pleading
- narration
- whining
- crying
- threatening
- raspy voice
- theatrical
- declamatory
- chanting
- weak voice
- voiceover