Hugging Face's logo

Datasets:
enactic
/
avsr-leaderboard

Tasks:
Automatic Speech Recognition
Languages:
Japanese
ArXiv:
Tags:
leaderboard
AVSR
License:
Dataset card Files
xet
 Community
yubo0306 commited on
Commit
5e1bae7
·
verified ·
1 Parent(s): 388415f

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +166 -0
README.md ADDED
@@ -0,0 +1,166 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - leaderboard
4
+ - AVSR
5
+ license: apache-2.0
6
+ task_categories:
7
+ - automatic-speech-recognition
8
+ language:
9
+ - ja
10
+ pretty_name: AVSR
11
+ ---
12
+
13
+ # Japanese AVSR Leaderboard
14
+
15
+ This is an AVSR leaderboard that evaluates AVSR/ASR models using an internally collected out-of-domain evaluation dataset for AVSR benchmarking.
16
+
17
+ ## Evaluation Dataset
18
+
19
+ We randomly sampled sentences from the [JSUT](https://sites.google.com/site/shinnosuketakamichi/publication/jsut) corpus, had about five speakers read them aloud while simultaneously recording their faces, and collected 660 audio-visual samples that passed manual quality checks.
20
+
21
+ ## Metrics
22
+
23
+ The model performance is evaluated using CER (Character Error Rate).
24
+ We also assess AVSR performance under noisy conditions by mixing the input speech with various types of background sounds — environmental noise, music, babble, the original [JSUT](https://sites.google.com/site/shinnosuketakamichi/publication/jsut) speech, and [ReazonSpeech](https://huggingface.co/datasets/reazon-research/reazonspeech) speech.
25
+ Furthermore, multiple noise levels are evaluated: no noise (N/A) and SNR = {-5, 0, 5, 10}, where 0 dB indicates equal power between speech and noise, and lower SNR values correspond to more noise-dominant audio.
26
+
27
+ ## Leaderboard
28
+
29
+ **Environmental Noise**
30
+
31
+ | Model | | #Params | N/A | SNR=10 | SNR=5 | SNR=0 | SNR=-5 |
32
+ | :------------------ | :----------------------------------------------------------------: | ------: | -----: | -----: | -----: | -----: | -----: |
33
+ | AVSR |||||||
34
+ | AVista Large+ v2 | [🤗](https://huggingface.co/enactic/avista-large-plus-v2) | 459M | 9.46% | 10.04% | 11.56% | 14.75% | 21.68% |
35
+ | AVista Large v2 | [🤗](https://huggingface.co/enactic/avista-large-v2) | 459M | 10.01% | 11.36% | 12.99% | 17.15% | 24.96% |
36
+ | AVista Base+ v2 | [🤗](https://huggingface.co/enactic/avista-base-plus-v2) | 156M | 10.77% | 12.70% | 14.99% | 21.10% | 31.99% |
37
+ | AVista Base v2 | [🤗](https://huggingface.co/enactic/avista-base-v2) | 156M | 11.11% | 12.81% | 14.96% | 20.58% | 31.09% |
38
+ | AVista Large+ | [🤗](https://huggingface.co/enactic/avista-large-plus) | 459M | 21.57% | 23.57% | 26.04% | 31.29% | 38.91% |
39
+ | AVista Large | [🤗](https://huggingface.co/enactic/avista-large) | 459M | 23.82% | 29.32% | 33.45% | 40.07% | 50.03% |
40
+ | AVista Base+ | [🤗](https://huggingface.co/enactic/avista-base-plus) | 156M | 26.88% | 28.65% | 30.60% | 35.43% | 44.16% |
41
+ | AVista Base | [🤗](https://huggingface.co/enactic/avista-base) | 156M | 27.46% | 31.34% | 34.80% | 41.53% | 49.72% |
42
+ | ASR ||||||
43
+ | reazonspeech k2 | [🤗](https://huggingface.co/reazon-research/reazonspeech-k2-v2) | 159M | 7.42% | 8.07% | 8.68% | 10.32% | 15.53% |
44
+ | reazonspeech nemo | [🤗](https://huggingface.co/reazon-research/reazonspeech-nemo-v2) | 619M | 8.50% | 9.31% | 10.16% | 12.71% | 18.32% |
45
+ | reazonspeech espnet | [🤗](https://huggingface.co/reazon-research/reazonspeech-espnet-v2) | 118M | 7.44% | 8.00% | 8.63% | 10.06% | 14.54% |
46
+ | whisper large-v3 | [🤗](https://huggingface.co/openai/whisper-large-v3) | 1,550M | 7.75% | 8.46% | 9.17% | 11.98% | 19.36% |
47
+ | whisper large-v2 | [🤗](https://huggingface.co/openai/whisper-large-v2) | 1,550M | 7.75% | 10.01% | 11.16% | 14.70% | 22.73% |
48
+ | whisper large-v1 | [🤗](https://huggingface.co/openai/whisper-large) | 1,550M | 8.93% | 10.01% | 11.87% | 15.48% | 23.34% |
49
+ | whisper medium | [🤗](https://huggingface.co/openai/whisper-medium) | 769M | 10.07% | 11.77% | 13.06% | 17.04% | 24.83% |
50
+ | whisper small | [🤗](https://huggingface.co/openai/whisper-small) | 244M | 10.82% | 17.62% | 19.84% | 25.55% | 33.77% |
51
+ | whisper base | [🤗](https://huggingface.co/openai/whisper-base) | 74M | 16.11% | 29.88% | 33.78% | 42.70% | 56.83% |
52
+ | whisper tiny | [🤗](https://huggingface.co/openai/whisper-tiny) | 39M | 26.45% | 43.78% | 49.48% | 59.69% | 75.30% |
53
+
54
+ **Music**
55
+
56
+ | Model | | #Params | N/A | SNR=10 | SNR=5 | SNR=0 | SNR=-5 |
57
+ | :------------------ | :----------------------------------------------------------------: | ------: | -----: | -----: | -----: | -----: | ------: |
58
+ | AVSR |||||||
59
+ | AVista Large+ v2 | [🤗](https://huggingface.co/enactic/avista-large-plus-v2) | 459M | 9.46% | 10.33% | 11.38% | 15.63% | 25.53% |
60
+ | AVista Large v2 | [🤗](https://huggingface.co/enactic/avista-large-v2) | 459M | 10.01% | 11.41% | 13.57% | 18.80% | 31.70% |
61
+ | AVista Base+ v2 | [🤗](https://huggingface.co/enactic/avista-base-plus-v2) | 156M | 10.77% | 12.51% | 14.81% | 21.85% | 40.73% |
62
+ | AVista Base v2 | [🤗](https://huggingface.co/enactic/avista-base-v2) | 156M | 11.11% | 12.23% | 14.84% | 21.55% | 39.76% |
63
+ | AVista Large+ | [🤗](https://huggingface.co/enactic/avista-large-plus) | 459M | 21.57% | 24.42% | 27.63% | 34.68% | 49.39% |
64
+ | AVista Large | [🤗](https://huggingface.co/enactic/avista-large) | 459M | 23.82% | 31.48% | 38.21% | 50.40% | 70.19% |
65
+ | AVista Base+ | [🤗](https://huggingface.co/enactic/avista-base-plus) | 156M | 26.88% | 27.91% | 31.29% | 41.01% | 56.38% |
66
+ | AVista Base | [🤗](https://huggingface.co/enactic/avista-base) | 156M | 27.46% | 31.74% | 37.15% | 49.03% | 65.76% |
67
+ | ASR ||||||
68
+ | reazonspeech k2 | [🤗](https://huggingface.co/reazon-research/reazonspeech-k2-v2) | 159M | 7.42% | 7.69% | 8.33% | 9.49% | 16.90% |
69
+ | reazonspeech nemo | [🤗](https://huggingface.co/reazon-research/reazonspeech-nemo-v2) | 619M | 8.50% | 9.28% | 9.97% | 13.65% | 24.61% |
70
+ | reazonspeech espnet | [🤗](https://huggingface.co/reazon-research/reazonspeech-espnet-v2) | 118M | 7.44% | 7.86% | 8.57% | 10.41% | 16.62% |
71
+ | whisper large-v3 | [🤗](https://huggingface.co/openai/whisper-large-v3) | 1,550M | 7.75% | 8.16% | 9.01% | 11.23% | 21.26% |
72
+ | whisper large-v2 | [🤗](https://huggingface.co/openai/whisper-large-v2) | 1,550M | 7.75% | 9.64% | 10.68% | 14.79% | 27.53% |
73
+ | whisper large-v1 | [🤗](https://huggingface.co/openai/whisper-large) | 1,550M | 8.93% | 10.50% | 11.86% | 15.77% | 27.52% |
74
+ | whisper medium | [🤗](https://huggingface.co/openai/whisper-medium) | 769M | 10.07% | 11.13% | 12.97% | 16.45% | 31.62% |
75
+ | whisper small | [🤗](https://huggingface.co/openai/whisper-small) | 244M | 10.82% | 18.02% | 19.86% | 26.82% | 47.69% |
76
+ | whisper base | [🤗](https://huggingface.co/openai/whisper-base) | 74M | 16.11% | 31.31% | 33.72% | 45.52% | 80.11% |
77
+ | whisper tiny | [🤗](https://huggingface.co/openai/whisper-tiny) | 39M | 26.45% | 41.37% | 49.06% | 66.81% | 104.54% |
78
+
79
+ **Babble**
80
+
81
+ | Model | | #Params | N/A | SNR=10 | SNR=5 | SNR=0 | SNR=-5 |
82
+ | :------------------ | :----------------------------------------------------------------: | ------: | -----: | -----: | -----: | ------: | ------: |
83
+ | AVSR |||||||
84
+ | AVista Large+ v2 | [🤗](https://huggingface.co/enactic/avista-large-plus-v2) | 459M | 9.46% | 11.09% | 15.05% | 26.94% | 61.66% |
85
+ | AVista Large v2 | [🤗](https://huggingface.co/enactic/avista-large-v2) | 459M | 10.01% | 12.52% | 17.78% | 32.06% | 66.75% |
86
+ | AVista Base+ v2 | [🤗](https://huggingface.co/enactic/avista-base-plus-v2) | 156M | 10.77% | 13.86% | 20.38% | 43.51% | 85.12% |
87
+ | AVista Base v2 | [🤗](https://huggingface.co/enactic/avista-base-v2) | 156M | 11.11% | 13.75% | 20.53% | 43.15% | 85.09% |
88
+ | AVista Large+ | [🤗](https://huggingface.co/enactic/avista-large-plus) | 459M | 21.57% | 26.85% | 33.91% | 51.24% | 79.48% |
89
+ | AVista Large | [🤗](https://huggingface.co/enactic/avista-large) | 459M | 23.82% | 32.19% | 42.46% | 61.45% | 89.13% |
90
+ | AVista Base+ | [🤗](https://huggingface.co/enactic/avista-base-plus) | 156M | 26.88% | 30.02% | 36.83% | 54.02% | 82.00% |
91
+ | AVista Base | [🤗](https://huggingface.co/enactic/avista-base) | 156M | 27.46% | 31.93% | 41.20% | 59.15% | 86.28% |
92
+ | ASR ||||||
93
+ | reazonspeech k2 | [🤗](https://huggingface.co/reazon-research/reazonspeech-k2-v2) | 159M | 7.42% | 8.24% | 10.17% | 21.65% | 61.57% |
94
+ | reazonspeech nemo | [🤗](https://huggingface.co/reazon-research/reazonspeech-nemo-v2) | 619M | 8.50% | 10.40% | 14.83% | 31.74% | 77.29% |
95
+ | reazonspeech espnet | [🤗](https://huggingface.co/reazon-research/reazonspeech-espnet-v2) | 118M | 7.44% | 8.85% | 11.75% | 24.59% | 67.27% |
96
+ | whisper large-v3 | [🤗](https://huggingface.co/openai/whisper-large-v3) | 1,550M | 7.75% | 8.95% | 12.50% | 30.09% | 81.60% |
97
+ | whisper large-v2 | [🤗](https://huggingface.co/openai/whisper-large-v2) | 1,550M | 7.75% | 10.68% | 15.73% | 36.95% | 98.40% |
98
+ | whisper large-v1 | [🤗](https://huggingface.co/openai/whisper-large) | 1,550M | 8.93% | 11.60% | 16.41% | 39.15% | 91.81% |
99
+ | whisper medium | [🤗](https://huggingface.co/openai/whisper-medium) | 769M | 10.07% | 12.52% | 18.18% | 42.27% | 95.43% |
100
+ | whisper small | [🤗](https://huggingface.co/openai/whisper-small) | 244M | 10.82% | 19.72% | 28.24% | 56.72% | 109.61% |
101
+ | whisper base | [🤗](https://huggingface.co/openai/whisper-base) | 74M | 16.11% | 32.84% | 48.84% | 92.20% | 220.22% |
102
+ | whisper tiny | [🤗](https://huggingface.co/openai/whisper-tiny) | 39M | 26.45% | 46.59% | 70.93% | 114.57% | 129.19% |
103
+
104
+ **JSUT Speech**
105
+
106
+ | Model | | #Params | N/A | SNR=10 | SNR=5 | SNR=0 | SNR=-5 |
107
+ | :------------------ | :----------------------------------------------------------------: | ------: | -----: | -----: | -----: | ------: | ------: |
108
+ | AVSR |||||||
109
+ | AVista Large+ v2 | [🤗](https://huggingface.co/enactic/avista-large-plus-v2) | 459M | 9.46% | 11.25% | 13.02% | 15.60% | 21.10% |
110
+ | AVista Large v2 | [🤗](https://huggingface.co/enactic/avista-large-v2) | 459M | 10.01% | 13.16% | 16.24% | 21.14% | 28.41% |
111
+ | AVista Base+ v2 | [🤗](https://huggingface.co/enactic/avista-base-plus-v2) | 156M | 10.77% | 14.49% | 21.41% | 57.51% | 91.28% |
112
+ | AVista Base v2 | [🤗](https://huggingface.co/enactic/avista-base-v2) | 156M | 11.11% | 14.90% | 22.89% | 62.42% | 91.61% |
113
+ | AVista Large+ | [🤗](https://huggingface.co/enactic/avista-large-plus) | 459M | 21.57% | 26.26% | 28.21% | 30.92% | 36.89% |
114
+ | AVista Large | [🤗](https://huggingface.co/enactic/avista-large) | 459M | 23.82% | 32.90% | 39.23% | 48.83% | 59.46% |
115
+ | AVista Base+ | [🤗](https://huggingface.co/enactic/avista-base-plus) | 156M | 26.88% | 31.56% | 34.03% | 38.85% | 47.72% |
116
+ | AVista Base | [🤗](https://huggingface.co/enactic/avista-base) | 156M | 27.46% | 34.30% | 39.15% | 45.85% | 55.92% |
117
+ | ASR ||||||
118
+ | reazonspeech k2 | [🤗](https://huggingface.co/reazon-research/reazonspeech-k2-v2) | 159M | 7.42% | 8.49% | 21.94% | 70.81% | 93.04% |
119
+ | reazonspeech nemo | [🤗](https://huggingface.co/reazon-research/reazonspeech-nemo-v2) | 619M | 8.50% | 10.93% | 29.06% | 83.77% | 98.76% |
120
+ | reazonspeech espnet | [🤗](https://huggingface.co/reazon-research/reazonspeech-espnet-v2) | 118M | 7.44% | 8.30% | 14.45% | 66.15% | 69.34% |
121
+ | whisper large-v3 | [🤗](https://huggingface.co/openai/whisper-large-v3) | 1,550M | 7.75% | 8.69% | 13.03% | 60.24% | 98.67% |
122
+ | whisper large-v2 | [🤗](https://huggingface.co/openai/whisper-large-v2) | 1,550M | 7.75% | 10.94% | 17.37% | 65.49% | 96.58% |
123
+ | whisper large-v1 | [🤗](https://huggingface.co/openai/whisper-large) | 1,550M | 8.93% | 11.31% | 15.86% | 60.68% | 97.25% |
124
+ | whisper medium | [🤗](https://huggingface.co/openai/whisper-medium) | 769M | 10.07% | 12.27% | 18.80% | 58.00% | 97.35% |
125
+ | whisper small | [🤗](https://huggingface.co/openai/whisper-small) | 244M | 10.82% | 19.44% | 26.75% | 71.33% | 101.84% |
126
+ | whisper base | [🤗](https://huggingface.co/openai/whisper-base) | 74M | 16.11% | 33.01% | 52.35% | 88.21% | 102.39% |
127
+ | whisper tiny | [🤗](https://huggingface.co/openai/whisper-tiny) | 39M | 26.45% | 51.68% | 79.56% | 102.10% | 107.11% |
128
+
129
+ **ReazonSpeech Speech**
130
+
131
+ | Model | | #Params | N/A | SNR=10 | SNR=5 | SNR=0 | SNR=-5 |
132
+ | :------------------ | :----------------------------------------------------------------: | ------: | -----: | -----: | -----: | ------: | ------: |
133
+ | AVSR |||||||
134
+ | AVista Large+ v2 | [🤗](https://huggingface.co/enactic/avista-large-plus-v2) | 459M | 9.46% | 11.77% | 14.84% | 21.41% | 34.31% |
135
+ | AVista Large v2 | [🤗](https://huggingface.co/enactic/avista-large-v2) | 459M | 10.01% | 13.71% | 17.38% | 25.48% | 41.32% |
136
+ | AVista Base+ v2 | [🤗](https://huggingface.co/enactic/avista-base-plus-v2) | 156M | 10.77% | 15.73% | 23.08% | 62.02% | 98.19% |
137
+ | AVista Base v2 | [🤗](https://huggingface.co/enactic/avista-base-v2) | 156M | 11.11% | 15.80% | 24.17% | 63.92% | 98.55% |
138
+ | AVista Large+ | [🤗](https://huggingface.co/enactic/avista-large-plus) | 459M | 21.57% | 27.73% | 31.36% | 39.61% | 54.82% |
139
+ | AVista Large | [🤗](https://huggingface.co/enactic/avista-large) | 459M | 23.82% | 35.18% | 44.70% | 59.01% | 78.08% |
140
+ | AVista Base+ | [🤗](https://huggingface.co/enactic/avista-base-plus) | 156M | 26.88% | 33.24% | 38.13% | 47.64% | 63.60% |
141
+ | AVista Base | [🤗](https://huggingface.co/enactic/avista-base) | 156M | 27.46% | 35.84% | 43.17% | 54.74% | 72.09% |
142
+ | ASR ||||||
143
+ | reazonspeech k2 | [🤗](https://huggingface.co/reazon-research/reazonspeech-k2-v2) | 159M | 7.42% | 9.13% | 19.47% | 71.61% | 104.15% |
144
+ | reazonspeech nemo | [🤗](https://huggingface.co/reazon-research/reazonspeech-nemo-v2) | 619M | 8.50% | 11.74% | 25.38% | 77.65% | 103.42% |
145
+ | reazonspeech espnet | [🤗](https://huggingface.co/reazon-research/reazonspeech-espnet-v2) | 118M | 7.44% | 9.20% | 16.58% | 69.34% | 103.22% |
146
+ | whisper large-v3 | [🤗](https://huggingface.co/openai/whisper-large-v3) | 1,550M | 7.75% | 8.70% | 12.81% | 49.34% | 100.53% |
147
+ | whisper large-v2 | [🤗](https://huggingface.co/openai/whisper-large-v2) | 1,550M | 7.75% | 11.10% | 18.11% | 56.80% | 106.04% |
148
+ | whisper large-v1 | [🤗](https://huggingface.co/openai/whisper-large) | 1,550M | 7.75% | 11.83% | 17.42% | 47.93% | 97.35% |
149
+ | whisper medium | [🤗](https://huggingface.co/openai/whisper-medium) | 769M | 10.07% | 13.23% | 19.21% | 50.56% | 99.27% |
150
+ | whisper small | [🤗](https://huggingface.co/openai/whisper-small) | 244M | 10.82% | 19.82% | 28.98% | 69.69% | 108.56% |
151
+ | whisper base | [🤗](https://huggingface.co/openai/whisper-base) | 74M | 16.11% | 35.53% | 54.26% | 97.67% | 115.63% |
152
+ | whisper tiny | [🤗](https://huggingface.co/openai/whisper-tiny) | 39M | 26.45% | 53.28% | 81.50% | 113.98% | 124.90% |
153
+
154
+ ## Acknowledgement
155
+
156
+ ```bibtex
157
+ @misc{sonobe2017jsutcorpusfreelargescale,
158
+ title={JSUT corpus: free large-scale Japanese speech corpus for end-to-end speech synthesis},
159
+ author={Ryosuke Sonobe and Shinnosuke Takamichi and Hiroshi Saruwatari},
160
+ year={2017},
161
+ eprint={1711.00354},
162
+ archivePrefix={arXiv},
163
+ primaryClass={cs.CL},
164
+ url={https://arxiv.org/abs/1711.00354},
165
+ }
166
+ ```