— For the latest version, download bark revival from github.
An adaptation (creative hack) of Bark TTS (GitHub)
This research explores the application of language models and machine-learning techniques to generate musical voices with personality. By utilising autoregressive transformers, specifically the Bark model, tokens are generated from input text to produce a unique, invented, and undefinable language. Composition is being made between text and sound using various control techniques, including token repetition and windowing, Lempel-Ziv-Welch compression, and token clustering from acoustic feature extraction, to regulate the granularity, intelligibility, and meaning of the output voice. A recursive generation system is also introduced, allowing for the creation of a large series of interrelated voices. The research is used in various artistic applications, including music remixing and theatre production. It explores other forms of expressive voices and storytelling, seamlessly lying right in the middle between text and sound.
Please, read: Musical Voice Synthesis at the Midpoint: Where Text Meets Sound (en, de, fr, jp)
Features:
- Experiments with code and audio examples.
- Explores application of language models and machine-learning techniques to generate musical voices with personality using autoregressive transformers (Bark model).
- Generates unique invented language from input text tokens.
- Composition techniques:
- Text-to-sound control: token repetition, windowing, Lempel-Ziv-Welch compression, token clustering from acoustic feature extraction.
- Recursive generation system for creating interrelated voices.
Applications:
- Music remixing.
- Theater-Music production.
- Other expressive voices and storytelling forms seamlessly blend text and sound.
Compatibility:
Windows 10 & 11 and MacOS 15 & 26