Bark: Suno releases an open text-to-audio model with laughter, sighs and music
In one sentence Suno releases Bark, a transformer-based text-to-audio model: besides multilingual speech it generates laughter, sighs, music and other nonverbal sounds marked with tags in the text. Since May 1, 2023 the license is MIT, so it can also be used commercially.
Traditional TTS systems read text flatly. Bark, released by Suno on April 20, 2023, instead generates audio straight from text, including nonverbal communication.
By putting tags such as "[laughs]", "[sighs]", "[gasps]", "[clears throat]" or "[music]" in the text, or the ♪ symbol for sung parts, Bark tries to actually produce those sounds. It supports several languages and automatically detects the language of the input text.
Code and weights are public (weights on Hugging Face). On May 1, 2023 the license switches to MIT, opening commercial use.
Companies
Suno AI
Tools
Bark, EnCodec
Tags
Sources