Bark
Bark is a transformer-based text-to-audio model created by Suno. Bark can generate highly realistic, multilingual speech as well as other audio - including music, background noise and simple sound effects. The model can also produce nonverbal communications like laughing, sighing and crying. To support the research community, we are providing access to pretrained model checkpoints ready for inference.
The original github repo and model card can be found here.
This model is meant for research purposes only. The model output is not censored and the authors do not endorse the opinions in the generated content. Use at your own risk.
The following is additional information about the models released here.
Model Usage
from bark import SAMPLE_RATE, generate_audio, preload_models
from IPython.display import Audio
# download and load all models
preload_models()
# generate audio from text
text_prompt = """
Hello, my name is Suno. And, uh — and I like pizza. [laughs]
But I also have other interests such as playing tic tac toe.
"""
audio_array = generate_audio(text_prompt)
# play text in notebook
Audio(audio_array, rate=SAMPLE_RATE)
To save audio_array as a WAV file:
from scipy.io.wavfile import write as write_wav
write_wav("/path/to/audio.wav", SAMPLE_RATE, audio_array)
Model Details
Bark is a series of three transformer models that turn text into audio.
Text to semantic tokens
- Input: text, tokenized with BERT tokenizer from Hugging Face
- Output: semantic tokens that encode the audio to be generated
Semantic to coarse tokens
- Input: semantic tokens
- Output: tokens from the first two codebooks of the EnCodec Codec from facebook
Coarse to fine tokens
- Input: the first two codebooks from EnCodec
- Output: 8 codebooks from EnCodec
Architecture
| Model | Parameters | Attention | Output Vocab size |
|---|---|---|---|
| Text to semantic tokens | 80/300 M | Causal | 10,000 |
| Semantic to coarse tokens | 80/300 M | Causal | 2x 1,024 |
| Coarse to fine tokens | 80/300 M | Non-causal | 6x 1,024 |
Release date
April 2023
Broader Implications
We anticipate that this model's text to audio capabilities can be used to improve accessbility tools in a variety of languages.
While we hope that this release will enable users to express their creativity and build applications that are a force for good, we acknowledge that any text to audio model has the potential for dual use. While it is not straightforward to voice clone known people with Bark, it can still be used for nefarious purposes. To further reduce the chances of unintended use of Bark, we also release a simple classifier to detect Bark-generated audio with high accuracy (see notebooks section of the main repository).
