Three Experiments on the Application of Automatic Speech Recognition in Industrial Environments

In this work we examine the performance of automatic speech recognition (ASR) in industrial applications. We particularly present three experiments relating to the capturing device applied, the signal preprocessing employed, and the recognition engine used. Here, our aim was to create experimental conditions as close as possible to the envisioned application, i.e., an industrial adoption of ASR. Our results show the existence of evident dependencies between the recognition engine, the type of capturing device, and the noise type on the one side, and the complexity of the task, the present SignaltoNoiseRatio (SNR), and the minimumacceptable SNR value on the other side. In summary, this work gives an overview of the capabilities and limitations of nowadays ASR systems for an application in an industrial context.