# SpeechBrain

An open-source PyTorch toolkit for conversational AI, covering speech recognition, speaker recognition, speech enhancement and separation, and language modeling, with more than 200 training recipes.

SpeechBrain is an open-source PyTorch toolkit built to accelerate development of conversational AI, the technology behind speech assistants, chatbots and large language models, by supporting speech and text processing together in one place. It covers speech recognition, speaker recognition, speech enhancement, speech separation, language modeling, dialogue and, more recently, an EEG modality aimed at non-verbal communication.

It shares more than 200 training recipes across more than 40 datasets, spanning about 20 speech and text-processing tasks, and supports both training a model from scratch and fine-tuning pretrained models such as Whisper, Wav2Vec2, WavLM, HuBERT, GPT-2 and Llama2, which can be pulled directly from Hugging Face. Training uses a YAML file for hyperparameters and a Python script to run the training loop, with a consistent code structure kept across different tasks.

More than 100 pretrained models are also available on Hugging Face, each with a simple interface; transcribing speech with a pretrained model, for example, takes only three lines of code.

Website: http://speechbrain.github.io