seastar105 commited on
Commit
1f130a4
Β·
verified Β·
1 Parent(s): eb0227a

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +45 -0
README.md ADDED
@@ -0,0 +1,45 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: transformers
3
+ tags: []
4
+ ---
5
+
6
+ ### Model Description
7
+
8
+ OpenAI의 whisper-base λͺ¨λΈμ„ μ•„λž˜ λ°μ΄ν„°μ…‹μœΌλ‘œ ν•™μŠ΅ν•œ λͺ¨λΈμž…λ‹ˆλ‹€. phonetic form을 μ‚¬μš©ν•˜μ—¬ ν•™μŠ΅λ˜μ—ˆμŠ΅λ‹ˆλ‹€.
9
+ - ν•œκ΅­μ–΄ μŒμ„± (https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=123)
10
+ - μ£Όμ†Œ μŒμ„± 데이터 (https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=71556)
11
+ - μ£Όμš” μ˜μ—­λ³„ 회의 μŒμ„±μΈμ‹ 데이터 (https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=464)
12
+ - μ €μŒμ§ˆ 전화망 μŒμ„±μΈμ‹ 데이터 (https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&dataSetSn=571)
13
+ - 방솑 μ½˜ν…μΈ  λŒ€ν™”μ²΄ μŒμ„±μΈμ‹ 데이터 (https://www.aihub.or.kr/aihubdata/data/view.do?dataSetSn=463)
14
+
15
+ ```
16
+ train_steps: 20000
17
+ warmup_steps: 2000
18
+ lr scheduler: linear warmup cosine decay
19
+ max learning rate: 1e-4
20
+ batch size: 256
21
+ max_grad_norm: 1.0
22
+ adamw_beta1: 0.9
23
+ adamw_beta2: 0.98
24
+ ```
25
+
26
+ ### Evaluation
27
+
28
+ https://github.com/rtzr/Awesome-Korean-Speech-Recognition
29
+
30
+ μœ„ λ ˆν¬μ§€ν† λ¦¬μ—μ„œ μ£Όμš” μ˜μ—­λ³„ 회의 μŒμ„±μ„ μ œμ™Έν•œ ν…ŒμŠ€νŠΈμ…‹ κ²°κ³Όμž…λ‹ˆλ‹€. μ•„λž˜ ν…Œμ΄λΈ”μ—μ„œ whisper_base_komixv2_phnκ°€ λ³Έ λͺ¨λΈ μ„±λŠ₯μž…λ‹ˆλ‹€.
31
+
32
+
33
+ | Model | cv_15_ko | fleurs_ko | kcall_testset | kconf_test | kcounsel_test | klec_testset | kspon_clean | kspon_other |
34
+ |-----------------------|----------|-----------|---------------|------------|---------------|--------------|-------------|-------------|
35
+ | whisper_base | 21.16 | 11.89 | 42.56 | 27.62 | 22.24 | 28.65 | 30.41 | 27.02 |
36
+ | whisper_base_komix | 15.42 | 7.16 | 20.86 | 14.24 | 12.64 | 13.44 | 12.26 | 12.12 |
37
+ | whisper_base_komixv2 | 13.97 | 9.29 | 10.58 | 14.85 | 12.52 | 16.01 | 14.23 | 14.93 |
38
+ | whisper_base_komixv2_phn | 12.81 | 8.27 | 9.5 | 13.26 | 11.33 | 14.24 | 13.11 | 13.3 |
39
+ | whisper_large_v3 | 5.11 | 3.72 | 5.45 | 9.35 | 3.83 | 8.46 | 15.08 | 12.89 |
40
+ | whisper_turbo | 5.38 | 3.95 | 5.89 | 9.77 | 4.21 | 9.27 | 16.49 | 13.54 |
41
+ | whisper_turbo_lora | 6.25 | 4.0 | 6.51 | 9.94 | 5.05 | 8.84 | 9.35 | 9.29 |
42
+
43
+ ### Acknowledgement
44
+ - λ³Έ λͺ¨λΈμ€ κ΅¬κΈ€μ˜ TRC ν”„λ‘œκ·Έλž¨μ˜ μ§€μ›μœΌλ‘œ ν•™μŠ΅ν–ˆμŠ΅λ‹ˆλ‹€.
45
+ - Research supported with Cloud TPUs from Google's TPU Research Cloud (TRC)