Update README.md
Browse filesreplace README with the corpus published README
README.md
CHANGED
@@ -60,3 +60,266 @@ configs:
|
|
60 |
- split: test
|
61 |
path: data/test-*
|
62 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
- split: test
|
61 |
path: data/test-*
|
62 |
---
|
63 |
+
# speechocean762: A non-native English corpus for pronunciation scoring task
|
64 |
+
|
65 |
+
## Introduction
|
66 |
+
Pronunciation scoring is a crucial technology in computer-assisted language learning (CALL) systems. The pronunciation quality scores might be given at phoneme-level, word-level, and sentence-level for a typical pronunciation scoring task.
|
67 |
+
|
68 |
+
This corpus aims to provide a free public dataset for the pronunciation scoring task.
|
69 |
+
Key features:
|
70 |
+
* It is available for free download for both commercial and non-commercial purposes.
|
71 |
+
* The speaker variety encompasses young children and adults.
|
72 |
+
* The manual annotations are in multiple aspects at sentence-level, word-level and phoneme-level.
|
73 |
+
|
74 |
+
This corpus consists of 5000 English sentences. All the speakers are non-native, and their mother tongue is Mandarin. Half of the speakers are Children, and the others are adults. The information of age and gender are provided.
|
75 |
+
|
76 |
+
Five experts made the scores. To avoid subjective bias, each expert scores independently under the same metric.
|
77 |
+
|
78 |
+
|
79 |
+
## The scoring metric
|
80 |
+
The experts score at three levels: phoneme-level, word-level, and sentence-level.
|
81 |
+
|
82 |
+
### Phoneme level
|
83 |
+
Score the pronunciation goodness of each phoneme within the words.
|
84 |
+
|
85 |
+
Score range: 0-2
|
86 |
+
* 2: pronunciation is correct
|
87 |
+
* 1: pronunciation is right but has a heavy accent
|
88 |
+
* 0: pronunciation is incorrect or missed
|
89 |
+
|
90 |
+
### Word level
|
91 |
+
Score the accuracy and stress of each word's pronunciation.
|
92 |
+
|
93 |
+
#### Accuracy
|
94 |
+
Score range: 0 - 10
|
95 |
+
* 10: The pronunciation of the word is perfect
|
96 |
+
* 7-9: Most phones in this word are pronounced correctly but have accents
|
97 |
+
* 4-6: Less than 30% of phones in this word are wrongly pronounced
|
98 |
+
* 2-3: More than 30% of phones in this word are wrongly pronounced. In another case, the word is mispronounced as some other word. For example, the student mispronounced the word "bag" as "bike"
|
99 |
+
* 1: The pronunciation is hard to distinguish
|
100 |
+
* 0: no voice
|
101 |
+
|
102 |
+
#### Stress
|
103 |
+
Score range: {5, 10}
|
104 |
+
* 10: The stress is correct, or this is a mono-syllable word
|
105 |
+
* 5: The stress is wrong
|
106 |
+
|
107 |
+
### Sentence level
|
108 |
+
Score the accuracy, fluency, completeness and prosodic at the sentence level.
|
109 |
+
|
110 |
+
#### Accuracy
|
111 |
+
Score range: 0 - 10
|
112 |
+
* 9-10: The overall pronunciation of the sentence is excellent, with accurate phonology and no obvious pronunciation mistakes
|
113 |
+
* 7-8: The overall pronunciation of the sentence is good, with a few pronunciation mistakes
|
114 |
+
* 5-6: The overall pronunciation of the sentence is understandable, with many pronunciation mistakes and accent, but it does not affect the understanding of basic meanings
|
115 |
+
* 3-4: Poor, clumsy and rigid pronunciation of the sentence as a whole, with serious pronunciation mistakes
|
116 |
+
* 0-2: Extremely poor pronunciation and only one or two words are recognizable
|
117 |
+
|
118 |
+
#### Completeness
|
119 |
+
Score range: 0.0 - 1.0
|
120 |
+
The percentage of the words with good pronunciation.
|
121 |
+
|
122 |
+
#### Fluency
|
123 |
+
Score range: 0 - 10
|
124 |
+
* 8-10: Fluent without noticeable pauses or stammering
|
125 |
+
* 6-7: Fluent in general, with a few pauses, repetition, and stammering
|
126 |
+
* 4-5: the speech is a little influent, with many pauses, repetition, and stammering
|
127 |
+
* 0-3: intermittent, very influent speech, with lots of pauses, repetition, and stammering
|
128 |
+
|
129 |
+
#### Prosodic
|
130 |
+
Score range: 0 - 10
|
131 |
+
* 9-10: Correct intonation at a stable speaking speed, speak with cadence, and can speak like a native
|
132 |
+
* 7-8: Nearly correct intonation at a stable speaking speed, nearly smooth and coherent, but with little stammering and few pauses
|
133 |
+
* 5-6: Unstable speech speed, many stammering and pauses with a poor sense of rhythm
|
134 |
+
* 3-4: Unstable speech speed, speak too fast or too slow, without the sense of rhythm
|
135 |
+
* 0-2: Poor intonation and lots of stammering and pauses, unable to read a complete sentence
|
136 |
+
|
137 |
+
## Data structure
|
138 |
+
The following tree shows the file structure of this corpus on [github](https://github.com/jimbozhang/speechocean762):
|
139 |
+
```
|
140 |
+
βββ scores.json
|
141 |
+
βββ scores-detail.json
|
142 |
+
βββ train
|
143 |
+
β βββ spk2age
|
144 |
+
β βββ spk2gender
|
145 |
+
β βββ spk2utt
|
146 |
+
β βββ text
|
147 |
+
β βββ utt2spk
|
148 |
+
β βββ wav.scp
|
149 |
+
βββ test
|
150 |
+
β βββ spk2age
|
151 |
+
β βββ spk2gender
|
152 |
+
β βββ spk2utt
|
153 |
+
β βββ text
|
154 |
+
β βββ utt2spk
|
155 |
+
β βββ wav.scp
|
156 |
+
βββ WAVE
|
157 |
+
βββ SPEAKER0001
|
158 |
+
β βββ 000010011.WAV
|
159 |
+
β βββ 000010035.WAV
|
160 |
+
β βββ ...
|
161 |
+
β βββ 000010173.WAV
|
162 |
+
βββ SPEAKER0003
|
163 |
+
β βββ 000030012.WAV
|
164 |
+
β βββ 000030024.WAV
|
165 |
+
β βββ ...
|
166 |
+
β βββ 000030175.WAV
|
167 |
+
βββ SPEAKER0005
|
168 |
+
βββ 000050003.WAV
|
169 |
+
βββ 000050010.WAV
|
170 |
+
βββ ...
|
171 |
+
βββ 000050175.WAV
|
172 |
+
```
|
173 |
+
|
174 |
+
There are two datasets: `train` and `test`, and both are in Kaldi's data directory style.
|
175 |
+
|
176 |
+
|
177 |
+
The scores are stored in `scores.json`. Here is an example:
|
178 |
+
|
179 |
+
```
|
180 |
+
{
|
181 |
+
"000010011": { # utt-id
|
182 |
+
"text": "WE CALL IT BEAR", # transcript text
|
183 |
+
"accuracy": 8, # sentence-level accuracy score
|
184 |
+
"completeness": 10.0, # sentence-level completeness score
|
185 |
+
"fluency": 9, # sentence-level fluency score
|
186 |
+
"prosodic": 9, # sentence-level prosodic score
|
187 |
+
"total": 8, # sentence-level total score
|
188 |
+
"words": [
|
189 |
+
{
|
190 |
+
"accuracy": 10, # word-level accuracy score
|
191 |
+
"stress": 10, # word-level stress score
|
192 |
+
"total": 10, # word-level total score
|
193 |
+
"text": "WE", # the word text
|
194 |
+
"phones": "W IY0", # phones of the word
|
195 |
+
"phones-accuracy": [2.0, 2.0] # phoneme-level accuracy score
|
196 |
+
},
|
197 |
+
{
|
198 |
+
"accuracy": 10,
|
199 |
+
"stress": 10,
|
200 |
+
"total": 10,
|
201 |
+
"text": "CALL",
|
202 |
+
"phones": "K AO0 L",
|
203 |
+
"phones-accuracy": [2.0, 1.8, 1.8]
|
204 |
+
},
|
205 |
+
{
|
206 |
+
"accuracy": 10,
|
207 |
+
"stress": 10,
|
208 |
+
"total": 10,
|
209 |
+
"text": "IT",
|
210 |
+
"phones": "IH0 T",
|
211 |
+
"phones-accuracy": [2.0, 2.0]
|
212 |
+
},
|
213 |
+
{
|
214 |
+
"accuracy": 6,
|
215 |
+
"stress": 10,
|
216 |
+
"total": 6,
|
217 |
+
"text": "BEAR",
|
218 |
+
"phones": "B EH0 R",
|
219 |
+
"phones-accuracy": [2.0, 1.0, 1.0]
|
220 |
+
}
|
221 |
+
]
|
222 |
+
},
|
223 |
+
...
|
224 |
+
}
|
225 |
+
```
|
226 |
+
|
227 |
+
For the phones with an accuracy score lower than 0.5, an extra "mispronunciations" block indicates which phoneme the current phone was actually pronounced.
|
228 |
+
An example:
|
229 |
+
|
230 |
+
```
|
231 |
+
{
|
232 |
+
"text": "LISA",
|
233 |
+
"accuracy": 5,
|
234 |
+
"phones": ["L", "IY1", "S", "AH0"],
|
235 |
+
"phones-accuracy": [0.4, 2, 2, 1.2],
|
236 |
+
"mispronunciations": [
|
237 |
+
{
|
238 |
+
"canonical-phone": "L",
|
239 |
+
"index": 0,
|
240 |
+
"pronounced-phone": "D"
|
241 |
+
}
|
242 |
+
],
|
243 |
+
"stress": 10,
|
244 |
+
"total": 6
|
245 |
+
}
|
246 |
+
```
|
247 |
+
|
248 |
+
The file `scores.json` is processed from `scores-detail.json`.
|
249 |
+
The two JSON files are almost the same, but `scores-detail.json` has the five experts' original scores, while the scores of scores.json were the average or median scores.
|
250 |
+
|
251 |
+
An example item in `scores-detail.json`:
|
252 |
+
```
|
253 |
+
{
|
254 |
+
"000010011": {
|
255 |
+
|
256 |
+
"text": "WE CALL IT BEAR",
|
257 |
+
"accuracy": [7.0, 9.0, 8.0, 8.0, 9.0],
|
258 |
+
"completeness": [1.0, 1.0, 1.0, 1.0, 1.0],
|
259 |
+
"fluency": [10.0, 9.0, 8.0, 8.0, 10.0],
|
260 |
+
"prosodic": [10.0, 9.0, 7.0, 8.0, 9.0],
|
261 |
+
"total": [7.6, 9.0, 7.9, 8.0, 9.1],
|
262 |
+
"words": [
|
263 |
+
{
|
264 |
+
"accuracy": [10.0, 10.0, 10.0, 10.0, 10.0],
|
265 |
+
"stress": [10.0, 10.0, 10.0, 10.0, 10.0],
|
266 |
+
"total": [10.0, 10.0, 10.0, 10.0, 10.0],
|
267 |
+
"text": "WE",
|
268 |
+
"ref-phones": "W IY0",
|
269 |
+
"phones": ["W IY0", "W IY0", "W IY0", "W IY0", "W IY0"]
|
270 |
+
},
|
271 |
+
{
|
272 |
+
"accuracy": [10.0, 8.0, 10.0, 10.0, 8.0],
|
273 |
+
"stress": [10.0, 10.0, 10.0, 10.0, 10.0],
|
274 |
+
"total": [10.0, 8.4, 10.0, 10.0, 8.4],
|
275 |
+
"text": "CALL",
|
276 |
+
"ref-phones": "K AO0 L",
|
277 |
+
"phones": ["K AO0 L", "K {AO0} L", "K AO0 L", "K AO0 L", "K AO0 {L}"],
|
278 |
+
},
|
279 |
+
{
|
280 |
+
"accuracy": [10.0, 10.0, 10.0, 10.0, 10.0],
|
281 |
+
"stress": [10.0, 10.0, 10.0, 10.0, 10.0],
|
282 |
+
"total": [10.0, 10.0, 10.0, 10.0, 10.0],
|
283 |
+
"text": "IT",
|
284 |
+
"ref-phones": "IH0 T",
|
285 |
+
"phones": ["IH0 T", "IH0 T", "IH0 T", "IH0 T", "IH0 T"]
|
286 |
+
},
|
287 |
+
{
|
288 |
+
"accuracy": [3.0, 7.0, 10.0, 2.0, 6.0],
|
289 |
+
"stress": [10.0, 10.0, 10.0, 10.0, 10.0],
|
290 |
+
"phones": ["B (EH0) (R)", "B {EH0} {R}", "B EH0 R", "B (EH0) (R)", "B EH0 [L] R"],
|
291 |
+
"total": [4.4, 7.6, 10.0, 3.6, 6.8],
|
292 |
+
"text": "BEAR",
|
293 |
+
"ref-phones": "B EH0 R"
|
294 |
+
}
|
295 |
+
],
|
296 |
+
},
|
297 |
+
...
|
298 |
+
}
|
299 |
+
```
|
300 |
+
|
301 |
+
In `scores-detail.json`, the phoneme-level scores are notated in the following convenient notation:
|
302 |
+
|
303 |
+
* for score 2, do not use any symbol
|
304 |
+
* for score 1, use "{}" symbol
|
305 |
+
* for score 0, use "()" symbol
|
306 |
+
* for the inserted phone, use the "[]" symbol
|
307 |
+
|
308 |
+
|
309 |
+
For example, "B (EH) R" means the score of EH is 0 while the scores of B and R are both 2,
|
310 |
+
"B EH [L] R" mean there is an unexpected phone "L" and the other phones are scored 2.
|
311 |
+
|
312 |
+
## Citation
|
313 |
+
Please cite our paper if you find this work useful:
|
314 |
+
|
315 |
+
```bibtex
|
316 |
+
@inproceedings{zhang2021speechocean762,
|
317 |
+
title={speechocean762: An Open-Source Non-native English Speech Corpus For Pronunciation Assessment},
|
318 |
+
author={Zhang, Junbo and Zhang, Zhiwen and Wang, Yongqing and Yan, Zhiyong and Song, Qiong and Huang, Yukai and Li, Ke and Povey, Daniel and Wang, Yujun},
|
319 |
+
booktitle={Proc. Interspeech 2021},
|
320 |
+
year={2021}
|
321 |
+
}
|
322 |
+
```
|
323 |
+
|
324 |
+
|
325 |
+
|