VoxCPM2

VoxCPM2 is a strong choice for developers who want an open-source voice cloning model without being tied to a monthly SaaS subscription. Its 30-language support, 48kHz output, voice design and two cloning approaches give it considerably more flexibility than a basic local TTS model.

The main drawback is the setup. This isn't a simple website where you upload audio, type a sentence and forget about the technical side. Local inference requires suitable hardware, and the project's own documentation says controllable cloning and voice design can vary between runs. For developers who value local control and open licensing, though, that trade-off is reasonable.

VoxCPM2 is an open-source speech model that creates multilingual speech, designs new voices and clones existing voices from reference audio. The latest 2B model supports 30 languages, voice design, controllable cloning and 48kHz audio output.

The model gives you three distinct ways to create speech. Voice Design creates a new voice from a text description, Controllable Cloning uses reference audio while allowing control over style and emotion, and Ultimate Cloning uses the reference recording and its transcript to preserve more of the speaker's vocal characteristics.

Unlike a typical voice-generation subscription, VoxCPM2 can be downloaded and run locally. The official project is released under Apache 2.0, and its current documentation reports real-time streaming performance on suitable NVIDIA hardware.