Text-to-Speech Synthesizer for Conlangs
Design a flexible text-to-speech (TTS) system to allow generation of synthetic speech for invented and constructed languages. The user provides a JSON file mapping of graphemes to specific phonetic sounds in IPA. The TTS system will synthesize the desired speech by concatenating short audio excepts of constitute phones matching the target phoneme, retrieved from a large database of sounds. The algorithm will include heuristic-based intelligence to select consecutive phones from a single source, where available, to help to increase the naturalness of the generated speech.
Objectives
- Provide a flexible user-interface to enable user to define custom phonological mappings.
- Harvest a dataset of phonetic sounds using the espeak-ng formant synthesizer.
- Design a TTS system that concatenates phones to produce synthesized speech.
- Write thorough Python documentation and create suites of unit tests.
- Provide documentation and a user guide.
- Deploy tool as a data science app using streamlit.
Motivations
While natural languages developed alongside human societies over centuries, constructed languages are those deliberately invented in modern times. Famous examples of constructed languages include Esperanto, Klingdon, and Quenya (Elvish). However, modern speech generation approaches rely on deep-learning and large datasets of spoken examples of the language. These methods are not appropriate for synthesizing most invented languages. We desire a tool to support hobbyist-linguists developing constructed languages. This tool allows the user to define a language's phonology and produce spoken examples in the given language.
Qualifications
Minimum Qualifications:
- interest in linguistics, phonology, or speech synthesis
- experience with Github and project management in Python
- experience with signal processing or natural language processing
- experience in linguistics
- knowledge of more than one natural language
Details
Project Partner:
Patrick Donnelly
NDA/IPA:No Agreement Required
Number Groups:1
Project Status:Accepting Applicants
Website:https://soundbendor.org/