A speech recognition device for dentistry includes a processor, and a memory storing one or more instructions. The instructions are executed by the processor to obtain a sound containing noise and speech generated during a dental treatment, perform a noise cancelling on the sound, and run a speech-to-text (STT) model trained using a self-supervised learning method. The STT model executes extracting a feature from each of the sound with the noise cancelling and from the sound without the noise cancelling, obtaining an encoding vector by assigning a weight to each of the features and processing the features, and obtaining a script for the speech by decoding the encoding vector. Further, the self-supervised learning method includes a fine-tuning process in which sounds containing noises and speeches that are generated during the dental treatment and a script for each of the speeches are used as a training data.