Inference

Inference is the stage where a trained AI model is run on new, real-world input to produce an output, such as separating a track or transcribing speech, as opposed to the training stage where the model learns. Inference speed and efficiency determine whether a model can run in real time or on-device. It is the step that happens every time the product is actually used.