Scenario, which is the motive for this Bachelor Thesis is using Pepper robot as a welcome manager on the Faculty of Electrical Engineering and Computing. More specifically, this Bachelor Thesis focuses on analysis and improvement of speech recognition of robot Pepper, which is used in the mentioned scenario. Two solutions were compared, inbuilt solution and Google Cloud Speech-to-Text service. Choregraphe program was used for testing the inbuilt solution, while audio files used for Google Cloud service are recorded using Choregraphe as well. Average confidence levels of the recognized words were measured for two different distances (0.40 m and 0.80 m), two different noise levels (47 dB and 80 dB) and different sentence lengths (1, 2, 3 and 4 or more words). Based on the total of 1200 sentences tested, it was concluded that the average confidence level is significantly higher when using Google Cloud services, although this conclusion should be taken with a dose of caution as it is not known how exactly the confidence levels are being calculated. Moreover, it was concluded that speech recognition quality does not depend on the noise level in the environment as much as it depends on the distance between the person and Pepper. Furthermore, the average confidence level of the recognized words does not necessarily decrease with the increase of the number of the words in a sentence, but it depends more on the frequency of words in the everyday use of language as well as on the correct pronunciation of the sentence. Finally, it is strongly suggested to implement Google Cloud services for high-quality speech recognition while using Pepper as a welcome manager at the Faculty.