Bark
Bark is an open-source, transformer-based text-to-audio model by Suno that generates realistic multilingual speech, music, sound effects, and nonverbal sounds like laughter. It solves the need for a fully generative audio model that can produce expressive and varied audio from text prompts, suitable for developers and researchers.
✨ Key features
- Generates realistic multilingual speech, music, and sound effects.
- Supports 13 languages out-of-the-box with automatic language detection.
- Includes 100+ speaker presets for voice variation.
- Can produce nonverbal sounds like laughter, sighs, and crying.
- Available for commercial use under MIT License.
- Works on CPU and GPU with options for smaller models.
🎯 Use cases
- Create voiceovers for videos or podcasts.
- Generate audio for games or interactive applications.
- Produce multilingual audio content without recording.
- Add sound effects or music to projects.
- Research in generative audio models.
📦 Installation
🧰 Requirements: Python 3.6+, PyTorch 2.0+, CUDA 11.7/12.0 for GPU, or CPU. Models are downloaded from Hugging Face.
pip install git+https://github.com/suno-ai/bark.git
or
git clone https://github.com/suno-ai/bark
cd bark && pip install .
🚀 Usage
from bark import SAMPLE_RATE, generate_audio, preload_models
from scipy.io.wavfile import write as write_wav
preload_models()
text_prompt = "Hello, my name is Suno. And, uh — and I like pizza. [laughs]"
audio_array = generate_audio(text_prompt)
write_wav("bark_generation.wav", SAMPLE_RATE, audio_array)
⚠️ Good to know
Bark is a generative model that may deviate from prompts, output length is limited to ~13-14 seconds, and it does not support custom voice cloning.
❓ FAQ
How do I specify where models are downloaded and cached?
Bark uses Hugging Face to download and store models. You can configure the cache location via Hugging Face environment variables like HF_HOME.
Why do Bark's generations sometimes differ from my prompts?
Bark is a GPT-style model, so it may take creative liberties, resulting in higher-variance outputs than traditional text-to-speech.
What voices are supported by Bark?
Bark supports 100+ speaker presets across supported languages. It does not currently support custom voice cloning.
Why is the output limited to ~13-14 seconds?
Bark's architecture and context window are optimized for generations of roughly that length.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.