<p>Passive soundproof earmuffs, which operate on the principle of physical sound insulation, are extensively utilized for hearing protection in noisy industrial environments. The integration of active noise control (ANC) technology into these earmuffs has enabled the design of active soundproof earmuffs that significantly reduce noise across the entire frequency spectrum, particularly in the low-frequency range. However, a common limitation of both passive and active soundproof earmuffs is their tendency to suppress speech signals along with ambient noise, making effective communication difficult for wearers in high-noise settings. This paper presents a speech enhancement approach for the active soundproof earmuffs system, premised on the use of a microphone array. In the redesigned system, strong ambient noise is first isolated by the passive structure and then further suppressed by the filtered-x least mean squares (FxLMS) algorithm. The microphone array captures speech signals mixed with target noise and they are processed synchronously using an improved generalized cross-correlation (GCC) delay estimation method that incorporates spectral subtraction. These synchronized microphone array signals are then fed into a typical generalized sidelobe canceller (GSC) for speech enhancement. The enhanced speech signals are integrated with the ANC subsystem, where they serve as an excitation signal for online secondary path identification. Simulations and experiments are conducted to validate the efficacy of the proposed speech enhancement method for the active soundproof earmuffs system.</p>