SpeechBrain is an open-source PyTorch-based speech toolkit designed as a holistic framework that mimics the human brain by jointly supporting diverse technologies for complex conversational AI systems. It supports over 16 speech and audio processing tasks including automatic speech recognition (ASR), speaker recognition and verification, speech separation and enhancement, text-to-speech synthesis, spoken language understanding, speaker diarization, emotion classification, and voice activity detection. The toolkit also handles text processing tasks like language modeling with transformer architectures and grapheme-to-phoneme conversion, and even EEG processing for brain-computer interfaces. SpeechBrain provides over 200 competitive training recipes across 40+ datasets, allowing users to train from scratch or fine-tune pretrained models from HuggingFace including Whisper, Wav2Vec2, and WavLM. Key infrastructure features include dynamic batching, mixed-precision training, multi-GPU support, and hyperparameter management via YAML configuration. With 11,300+ stars and active development, SpeechBrain has become a comprehensive foundation for speech AI research and production deployments.