A New Acoustic-Based Pronunciation Distance Measure.

Martijn Bartelds,Mark Liberman,Martijn Wieling,Caitlin Richter

doi:10.3389/frai.2020.00039

Martijn Bartelds, Mark Liberman + Show 2 more

Open Access

https://doi.org/10.3389/frai.2020.00039

Copy DOI

Abstract

We present an acoustic distance measure for comparing pronunciations, and apply the measure to assess foreign accent strength in American-English by comparing speech of non-native American-English speakers to a collection of native American-English speakers. An acoustic-only measure is valuable as it does not require the time-consuming and error-prone process of phonetically transcribing speech samples which is necessary for current edit distance-based approaches. We minimize speaker variability in the data set by employing speaker-based cepstral mean and variance normalization, and compute word-based acoustic distances using the dynamic time warping algorithm. Our results indicate a strong correlation of r = −0.71 (p < 0.0001) between the acoustic distances and human judgments of native-likeness provided by more than 1,100 native American-English raters. Therefore, the convenient acoustic measure performs only slightly lower than the state-of-the-art transcription-based performance of r = −0.77. We also report the results of several small experiments which show that the acoustic measure is not only sensitive to segmental differences, but also to intonational differences and durational differences. However, it is not immune to unwanted differences caused by using a different recording device.

Highlights

The strength of foreign accent in a second language is mainly caused by the first language background of non-native speakers, and is influenced by a wide variety of variables with the most valuable predictor being the age of second-language learning (Asher and García, 1969; Leather, 1983; Flege, 1988; Arslan and Hansen, 1997)
Understanding the factors that affect the degree of foreign accent may be essential for second language teaching, and knowledge about the acoustic features of foreign-accented speech can improve speech recognition models (Arslan and Hansen, 1996; Piske et al, 2001)
To assess whether language variation within the set of reference speakers might be important, we computed the acoustic distances using as our reference set (N = 14) only the native American-English speakers who originated from the western half of the U.S and the English-speaking part of Canada

Summary

Introduction

The strength of foreign accent in a second language is mainly caused by the first language background of non-native speakers, and is influenced by a wide variety of variables with the most valuable predictor being the age of second-language learning (Asher and García, 1969; Leather, 1983; Flege, 1988; Arslan and Hansen, 1997). Computational methods that investigate foreign accent strength are, scarce. Studies that investigate and compare different pronunciations often use transcribed speech (Nerbonne and Heeringa, 1997; Livescu and Glass, 2000; Gooskens and Heeringa, 2004; Heeringa, 2004; Wieling et al, 2011; Chen et al, 2016; Jeszenszky et al, 2017). To calculate the Levenshtein distance, speech samples have to be manually transcribed using a

Objectives

Methods

Results

Conclusion