As voice-activated embedded controllers rapidly spread into use across multiple devices in the IoT, wearables, and smart home environments, energy efficiency has become a pressing concern, given the inability to power them off and the limited battery capacity. Fully digital VLSI designs of traditional voice recognition systems typically consume a significant amount of power and therefore cannot be utilised in low-power embedded applications. Purely analogue solutions, on the other hand, though energy efficient, are characterised by degradation in accuracy and limited programmability, particularly in changing environmental noise conditions. To overcome these challenges, this paper presents a new energy-conscious hybrid digital-analogue VLSI architecture that combines the low-power opportunities of analogue signal processing with the flexibility and accuracy of digital computation. The design features an analogue front-end that performs real-time voice signal conditioning and feature extraction at low voltages, and a digital back-end that recognises robust voice commands at low voltages using lightweight classifiers. An adaptive power management program is used to switch processing between analogue and digital modes of operation in real-time, based on the signal quality and noise levels of the real-time signal, to optimally use energy without affecting recognition performance. Digitally assisted calibration circuits also correct analogue non-idealities, ensuring the circuits remain stable in the face of process and temperature changes. Simulations of experimental models based on standard voice data sets have shown that the proposed hybrid VLSI-based design can achieve up to 45% less power consumption than current state-of-the-art fully digital counterparts, and voice recognition can be achieved with over 85% accuracy in noisy conditions. These findings confirm that the suggested hybrid solution is an effective means of balancing energy efficiency and system performance, which is why it is a highly valid solution in next-generation always-on voice-activated embedded controllers. The scalability and adaptability of the architecture ensure a wide range of applications for various low-power voice interface solutions, as well as future applications of more sustainable, embodied voice solutions.