Bark is Suno's open-source text-to-audio model, and it is broader than a text-to-speech system. Given a text prompt it can produce speech, but also music, ambient sound and non-verbal noises such as laughter or sighs, which conventional TTS models simply cannot generate.
That breadth is what makes it interesting technically. Most speech models are trained to read text cleanly and treat everything else as noise to suppress; Bark treats audio generation as one problem, which produces more expressive and more variable output. Variability is the trade-off, since the same prompt does not always give the same result.
It is distributed as a model on GitHub rather than a hosted product, so using it means running it yourself with the GPU and Python setup that implies. Being open source, it can run entirely locally with no per-use cost or data leaving your machine, the usual reason to choose it over a commercial API.
