Wearable acoustic sensors can be used for voice recognition. However, the capabilities of such devices, which are typically based on solid materials, are often restricted by ambient noise, motion artefacts and low conformability to the skin. Here we report a liquid acoustic sensor for voice recognition. The approach is based on a three-dimensional oriented and ramified magnetic network structure of neodymium–iron–boron magnetic nanoparticles suspended in a carrier fluid, which behaves like a permanent magnet. The sensor can discriminate small pressures (0.9 Pa), has a high signal-to-noise ratio (69.1 dB) and provides self-filtering capabilities that can remove low-frequency biomechanical motion artefact (less than 30 Hz). We use the liquid acoustic sensor—together with a machine learning algorithm—to create a wearable voice recognition system that offers a recognition accuracy of 99% in a noisy environment.

可穿戴声学传感器可用于语音识别。然而，此类设备通常基于固体材料，其功能常常受到环境噪声、运动伪影和皮肤适应性差的限制。在这里，我们报告了一种用于语音识别的液体声学传感器。该方法基于悬浮在载液中的钕铁硼磁性纳米粒子的三维定向和分支磁网络结构，其行为类似于永磁体。该传感器可以区分小压力 (0.9 Pa)，具有高信噪比 (69.1 dB)，并提供自滤波功能，可以消除低频生物力学运动伪影（小于 30 Hz）。我们使用液体声学传感器与机器学习算法一起创建可穿戴语音识别系统，该系统在嘈杂的环境中提供 99% 的识别准确率。