This paper presents a novel approach to neuromorphic computing using nanoscale spintronic oscillators. Inspired by the rhythmic activity and information processing of neurons in the brain, the authors demonstrate that these oscillators can achieve spoken digit recognition with accuracies comparable to state-of-the-art neural networks. The key advantages of spintronic oscillators include their small size, low power consumption, and ability to interact with each other, making them suitable for large-scale neuromorphic computing. The oscillators are composed of two ferromagnetic layers separated by a non-magnetic spacer, and they generate sustained magnetization precession at frequencies ranging from megahertz to gigahertz. The amplitude dynamics of these oscillators, which are robust to noise and exhibit non-linear behavior, are exploited for neuromorphic computing. The authors show that by using a single oscillator to emulate a neural network, they can achieve high recognition rates for spoken digits, even with added noise due to their nanometric size. The optimal operating conditions for pattern recognition are determined, and the results highlight the importance of balancing low noise and large amplitude variations for efficient neuromorphic computing. This work opens the door to fast, parallel, on-chip computation based on networks of interacting nanoscale spintronic oscillators.This paper presents a novel approach to neuromorphic computing using nanoscale spintronic oscillators. Inspired by the rhythmic activity and information processing of neurons in the brain, the authors demonstrate that these oscillators can achieve spoken digit recognition with accuracies comparable to state-of-the-art neural networks. The key advantages of spintronic oscillators include their small size, low power consumption, and ability to interact with each other, making them suitable for large-scale neuromorphic computing. The oscillators are composed of two ferromagnetic layers separated by a non-magnetic spacer, and they generate sustained magnetization precession at frequencies ranging from megahertz to gigahertz. The amplitude dynamics of these oscillators, which are robust to noise and exhibit non-linear behavior, are exploited for neuromorphic computing. The authors show that by using a single oscillator to emulate a neural network, they can achieve high recognition rates for spoken digits, even with added noise due to their nanometric size. The optimal operating conditions for pattern recognition are determined, and the results highlight the importance of balancing low noise and large amplitude variations for efficient neuromorphic computing. This work opens the door to fast, parallel, on-chip computation based on networks of interacting nanoscale spintronic oscillators.