What it does
Bark is an open-source model from Suno that turns a text prompt into audio. It speaks a range of languages out of the box, detecting the language from the input text, and comes with more than 100 speaker presets. It can also produce non-speech sounds such as laughing or sighing, music when lyrics are wrapped in music notes, and simple sound effects. Suno describes Bark as a research-oriented generative model rather than a conventional speech system, so its output can drift from the prompt and varies in audio quality. It does not support custom voice cloning, and one generation is limited to roughly 13 to 14 seconds, with a notebook provided for longer audio. Bark is released under the MIT license, with pretrained checkpoints available for commercial use. You run it yourself from Python, from a command line, or through the Hugging Face Transformers library. The full models need about 12 GB of GPU memory, a smaller version fits in roughly 8 GB, and it also runs on CPU, more slowly.
Capabilities
- Generates speech in 13 listed languages and detects the language from the text
- More than 100 speaker presets, plus randomly generated voices
- Non-speech cues such as [laughs], [sighs] and [music] written into the prompt
- Can render music and simple sound effects as well as speech
- Pretrained checkpoints under the MIT license, usable commercially
- Smaller model option for GPUs with limited memory
When to use it
- Prototyping spoken audio that needs expressive extras such as laughter or hesitation
- Experimenting with multilingual speech using preset voices
- Running an open audio model on your own hardware in a Python project
