notesum.ai

Published at November 7

Dialectal Coverage And Generalization in Arabic Speech Recognition

cs.CL

cs.SD

eess.AS

Released Date: November 7, 2024

Authors: Amirbek Djanibekov¹, Hawau Olamide Toyin¹, Raghad Alshalan², Abdullah Alitr², Hanan Aldarmaki¹

Aff.: ¹Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE; ²STC, Riyadh, Saudi Arabia

Arxiv: http://arxiv.org/abs/2411.05872v1

Dataset	Zero-Shot				Fine-Tuning
Dataset	Whisper	MMS	v1	v2	v1	v2
Mixat	56.01	61.51	39.1	35.05	34.29	33.40
TARIC-SLU	138.14	93.54	107.56	106.46	14.70	14.80
ParallelCorp	99.17	83.16	128.72	141.92	9.57	9.31
SADA	82.16	78.28	39.41	29.77	39.24	29.91
MASC Al-Fetyani et al. (2021)
SAU	48.39	65.30	61.23	58.72	27.40	27.33
SYR	26.65	33.21	21.99	18.37	18.64	17.42
EGY	41.73	66.04	50.87	47.17	38.47	36.43
JOR	28.65	54.63	61.23	34.97	19.72	21.08
LEB	40.95	64.58	35.65	42.66	30.01	28.05
IRA	41.69	59.33	50.46	48.03	31.10	34.64
TUN	47.45	60.58	50.37	46.67	19.26	18.52
MOR	65.87	80.84	78.92	66.87	47.59	49.40
PAL	53.20	83.72	77.94	73.53	55.88	53.53
KUW	36.00	81.71	64.74	52.02	50.29	46.24