Investigating place names of multilingual origins on the performance of automatic speech recognition systems

Abstract

This paper looks at automatic speech recognition and its performance on recognizing and transcribing place names of multilingual origins in Singapore English. 90 place names from Malay, Chinese and English origins were tested in an isolated and continuous speech task manner on two commercial ASR systems – Apple’s Siri and Google’s Google Assistant. Results showed that Siri outperformed Google for all trials, and both ASR systems recognized English-origin place names the best, followed by Malay and lastly Chinese.