Suno Bark

Bark is an open-source, transformer-based text-to-audio model released by Suno on GitHub. Unlike a conventional text-to-speech system, Bark generates audio directly from text, including speech, music, background noise, and simple sound effects, and it can also produce non-verbal sounds like laughing, sighing, or crying when cued in the input text. The model follows a GPT-style architecture similar to AudioLM and VALL-E, using a quantized audio representation from EnCodec, and it supports speech generation with automatic language detection across languages including English, German, Spanish, French, Hindi, Italian, Japanese, Korean, Polish, Portuguese, Russian, Turkish, and simplified Chinese, with over 100 speaker presets available. Bark is released under the MIT license, making it free for commercial use, and can be installed via pip or used through Hugging Face Transformers (version 4.31.0 and up), with hosted demos on Hugging Face Spaces, Replicate, and Google Colab. Running the full model requires around 12GB of VRAM, though a smaller configuration can fit into 8GB of VRAM or run on CPU. Because it is a generative rather than deterministic model, its output can deviate unpredictably from the exact prompt given.