What Can an Accent Identifier Learn? Probing Phonetic and Prosodic Information in a Wav2vec2-based Accent Identification Model

Mu Yang¹, Ram C. M. C. Shekar¹, Okim Kang², John H. L. Hansen¹

¹Center for Robust Speech Systems (CRSS), University of Texas at Dallas, USA

²Northern Arizona University, USA

This page provides additional examples for our Interspeech 2023 paper: What Can an Accent Identifier Learn? Probing Phonetic and Prosodic Information in a Wav2vec2-based Accent Identification Model.

Click the grey buttons and scroll to show figures.

1. The full list of CCA and MSE scores on the 7 accents: AB, CN, IN, KR, SP, US, VN.

(Click on the buttons to toggle the figure)

plots/all_curves.png — CCA and MSE scores

2. t-SNE visualization for all ARPAbet phonemes.

(Click on the buttons to toggle the figures)

Vowels

plots/AA.png — AA

plots/AE.png — AE

plots/AH.png — AH

plots/AO.png — AO

plots/AW.png — AW

plots/AY.png — AY

plots/EH.png — EH

plots/ER.png — ER

plots/EY.png — EY

plots/IH.png — IH

plots/IY.png — IY

plots/OW.png — OW

plots/OY.png — OY

plots/UH.png — UH

plots/UW.png — UW

Stops

plots/B.png — B

plots/D.png — D

plots/G.png — G

plots/K.png — K

plots/P.png — P

plots/T.png — T

Fricatives

plots/DH.png — DH

plots/F.png — F

plots/S.png — S

plots/SH.png — SH

plots/TH.png — TH

plots/V.png — V

plots/Z.png — Z

plots/ZH.png — ZH

Affricates

plots/CH.png — CH

plots/JH.png — JH

Nasals

plots/M.png — M

plots/N.png — N

plots/NG.png — NG

Liquids

plots/L.png — L

plots/R.png — R

Glides

plots/W.png — W

plots/Y.png — Y

Silence

plots/sil.png — sil

3. Bonus: t-SNE visualization of multiple phonemes of a specific accent.

(Click on the buttons to toggle the figures)

plots_accent/AB.png — AB

plots_accent/CN.png — CN

plots_accent/IN.png — IN

plots_accent/KR.png — KR

plots_accent/SP.png — SP

plots_accent/US.png — US

plots_accent/VN.png — VN