Abstract

Computer-generated text or speech, e.g., as utilized by assistant applications and bots with voice capabilities, utilizes words and phrases that are chosen carefully by the underlying language generation functions. Such text/speech usually lacks the nuance of real-world human speech that often provides rich indications of the personality, sociocultural characteristics, feelings, and thoughts of a human speaker. The techniques of this disclosure are usable to generate text/speech that expresses personal characteristics akin to a human speaker, and support representing a wide variety of personalities. The techniques involve preparing speech models of human characters by analyzing language use in existing conversation texts to cluster speech with similar word choices and speech patterns into character templates. When an assistant application or bot delivers output via text or voice, an appropriate character is chosen from among the templates. The characteristics of the template are applied to revise the originally generated response content to fit the character template.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS