Odyssey 2018 - Segmental Content Effects on Text-dependent Automatic Accent Recognition

Speaker Recognition

Geogina Brown

This paper investigates the effects of an unknown speech sample’s segmental content (the specific vowels and consonants it contains) on its chances of being successfully classified by an automatic accent recognition system. While there has been some work to investigate this effect in automatic speaker recognition, it has not been explored in relation to automatic accent recognition. This is a task where we would hypothesise that segmental content has a particularly large effect on the likelihood of a successful classification, especially for shorter speech samples. By focussing on one particular text-dependent automatic accent recognition system, the Y-ACCDIST system, we uncover the phonemes that appear to contribute more or less to successful classifications using a corpus of Northern English accents. We also relate these findings to the sociophonetic literature on these specific spoken varieties to attempt to account for the patterns that we see and to consider other factors that might contribute to a sample’s successful classification.

Cite as: Brown, G. (2018) Segmental Content Effects on Text-dependent Automatic Accent Recognition . Proc. Odyssey 2018 The Speaker and Language Recognition Workshop, 9-15, DOI: 10.21437/Speaker Odyssey.2018-2.