Whisper is OpenAI's family of speech recognition models using transformer encoder-decoder architecture, ranging from 39M to 1.55B parameters, capable of robust multilingual transcription, translation, and language identification across diverse acoustic conditions.

AudioCraft is Meta's suite of AI models for audio/music generation, featuring MusicGen and MAGNeT. Built on EnCodec compression and T5 encoding, these models generate high-quality 32kHz audio through transformers and hybrid architectures.