MeloTTS

MeloTTS came out of a collaboration between MIT and MyShell.ai as a multilingual text-to-speech library with an unusual emphasis on CPU speed: it performs real-time inference without a GPU. Language coverage includes English with American, British, Indian, and Australian accents, plus Spanish, French, Japanese, Korean, and Chinese, including mixed Chinese and English sentences. The implementation builds on the VITS family of architectures (VITS, VITS2, and Bert-VITS2) and can be used through a Python API, a command line interface, a web UI, or hosted demos, with documentation for training voices on custom datasets. Everything ships under the MIT license, which allows free commercial and non-commercial use, a point that distinguishes it from several popular TTS projects with restrictive terms. It is a frequent choice for embedding speech output in applications, voice agents, and edge deployments where GPU hosting is impractical.