EveryVoiceTTS/EveryVoice

Built on PyTorch and trained via FastSpeech2 architecture, it supports multi-speaker synthesis and handles low-resource language scenarios with configurable acoustic and vocoder models. The toolkit integrates sox and ffmpeg for audio preprocessing, uses YAML-based configuration schemas for reproducible training pipelines, and targets indigenous and endangered language communities through structured training workflows documented in its comprehensive guide system.

Available on PyPI.