Bark

Bark, from Suno, is a transformer-based text-to-audio model that goes beyond conventional TTS: alongside realistic multilingual speech it can generate music, background noise, and simple sound effects, plus nonverbal cues like laughter, sighs, and gasps triggered by inline tags such as [laughter] or a music-note marker. Thirteen languages are fully supported, including English, German, Spanish, French, Hindi, Japanese, Korean, Polish, Russian, and simplified Chinese, and the model handles code-switched text with appropriate accents while offering more than 100 speaker presets. Individual generations run about 13 to 14 seconds, with community recipes for long-form audio. The full model wants roughly 12GB of VRAM, while smaller variants and CPU offloading bring requirements down to 8GB or even 2GB systems. Bark is MIT licensed, permitting commercial use, and is available through the GitHub repository, Hugging Face Transformers, Spaces demos, Replicate, and Google Colab.