Whisper Perf - Public 2022-09-27

	A	B	C	D	E	F	G	H	I	J	K	M	N	P	Q	S
1	Dataset (Test)	Whisper Model WER										Talon Model WER		NVIDIA Model WER		This data shows the Word Error Rate of each model per dataset. Lower is better.
2		tiny	tiny.en	base	base.en	small	small.en	medium	med.en	large	large-v2	c-300M	d-1B	large	xlarge	(Check out the other tabs at the bottom)

3	common voice v10	29.27	26.76	21.64	19.75	14.46	14.18	11.17	11.66	10.22	9.63	13.61	9.44	7.33	5.78
4	librispeech clean	6.95	5.68	5.09	4.70	3.61	3.12	2.87	2.81	2.80	2.72	2.59	2.45	1.64	1.49
5	librispeech other	15.78	13.42	11.60	10.49	7.67	7.19	5.95	5.88	5.46	5.26	6.33	5.46	3.52	2.82
6	mls	17.22	15.17	13.02	12.01	9.25	9.04	8.24	7.92	7.43	7.85	9.37	8.04	5.59	5.14
7	gigaspeech	14.81	13.12	12.50	11.99	11.17	10.84	10.54	10.52	10.60	11.41	16.53	15.30	12.51	11.98
8	tedlium	9.95	8.24	7.68	7.19	7.40	6.65	6.68	6.23	6.97	8.47	6.42	5.31	6.76	6.28
9	accent2-all	18.19	17.25	13.94	13.05	9.09	9.36	7.56	7.82	7.39	6.12	12.49	10.45	11.50	8.54
10	podcast1 (n)	8.35	7.05	6.64	7.20	5.46	5.46	5.04	4.97	5.28	7.10	4.80	3.92	6.12	5.93
11	english1 (r)	11.27	9.63	9.20	8.88	6.94	6.47	6.18	6.26	6.09	6.87	8.22	7.56	7.36	7.16
12	podcast2 (t)	25.05	22.68	22.20	21.01	19.88	19.13	19.68	18.29	18.62	21.65	13.92	13.12	21.51	20.84
13	tts1	56.11	51.10	47.70	45.48	28.55	31.91	24.11	26.51	22.44	25.34	4.85	4.81	27.78	33.82
14	tts2	102.97	100.98	85.32	90.11	67.38	61.26	59.31	55.08	54.52	54.32	25.84	24.75	78.41	74.63
15	words1	51.97	45.02	35.16	30.74	24.06	21.69	18.85	18.07	17.82	16.70	8.05	7.55	30.61	31.85
16	words2	170.34	167.19	139.57	140.30	115.52	107.12	96.37	97.65	91.97	88.11	21.55	22.26	131.94	136.06
17	words3	61.37	54.41	48.28	45.85	35.39	33.33	28.76	27.51	25.53	26.23	16.14	16.16	32.37	28.92
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100