Datasets:
small fix in json file
Browse files
README.md
CHANGED
@@ -63,30 +63,6 @@ from datasets import load_dataset
|
|
63 |
data = load_dataset("nb/NPSC", streaming=True)
|
64 |
```
|
65 |
|
66 |
-
## Dataset Summary
|
67 |
-
The NPSC dataset contains json lines with language training data. Here is an example json line:
|
68 |
-
```json
|
69 |
-
|
70 |
-
{
|
71 |
-
"sentence_id": 49853,
|
72 |
-
"sentence_order": 0,
|
73 |
-
"speaker_id": 32,
|
74 |
-
"speaker_name": "Olemic Thommessen",
|
75 |
-
"sentence_text": "Stortingets møte er lovlig satt",
|
76 |
-
"sentence_language_code": "nb-NO",
|
77 |
-
"text": "Stortingets møte er lovlig satt",
|
78 |
-
"start_time": 320246, "end_time": 323590,
|
79 |
-
"normsentence_text": "Stortingets møte er lovlig satt",
|
80 |
-
"transsentence_text": "Stortingets møte er lovleg sett",
|
81 |
-
"translated": 1,
|
82 |
-
"audio": {"path": "audio/20170110-095504_320246_323590.wav",
|
83 |
-
"array": [.......]
|
84 |
-
}
|
85 |
-
|
86 |
-
}
|
87 |
-
|
88 |
-
|
89 |
-
```
|
90 |
## Data Fields
|
91 |
Currently there are two versions included in this repo.
|
92 |
|
@@ -95,25 +71,63 @@ This verison has a short list of the metadata and includes the audio (48k mp3) e
|
|
95 |
|
96 |
The current dataloader script is associated with this version.
|
97 |
|
98 |
-
|
99 |
-
|:-----------|:------------|
|
100 |
-
|**sentence_order** | String with order of sentence |
|
101 |
-
|**speaker id** | Integer id of speaker |
|
102 |
-
| **speaker_name** | String name of speaker |
|
103 |
-
| **sentence_text** | String sentence text |
|
104 |
-
| **sentence_language_code** | String sentence text |
|
105 |
-
| **text** | String sentence text |
|
106 |
-
| **start_time** | int start time |
|
107 |
-
| **end_time** | int end time |
|
108 |
-
| **normsentence_text** | String normalised sentence text |
|
109 |
-
| **transsentence_text** | String translated sentence text |
|
110 |
-
| **translated** | int text translated |
|
111 |
-
| **audio** | audio audio record with 'path',(mp3) 'array','sampling_rate' (48000) |
|
112 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
|
114 |
### Version B
|
115 |
This verison does not contain the audio encoded in the dataset. Instead it has the audio files placed in sub-directories. There are currently both samples in clips_48k_wav and clips_16k_mp3. Only the base filename is referred in the dataset. Please not that there are both sentence-based audio clips as well at meeting-based audio clips. The dataset contains referrals to both, the latter referral has start and stop time as well.
|
116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
|
118 |
### Dataset Creation
|
119 |
We are providing a **train**, **dev** and **test** split. These are the same as in the orginal corpus.
|
|
|
63 |
data = load_dataset("nb/NPSC", streaming=True)
|
64 |
```
|
65 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
## Data Fields
|
67 |
Currently there are two versions included in this repo.
|
68 |
|
|
|
71 |
|
72 |
The current dataloader script is associated with this version.
|
73 |
|
74 |
+
One line in train.json looks like this:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
+
```json
|
77 |
+
{
|
78 |
+
"sentence_id": 7309,
|
79 |
+
"sentence_order": 0,
|
80 |
+
"speaker_id": 1,
|
81 |
+
"speaker_name": "Marit Nybakk",
|
82 |
+
"sentence_text": "Stortingets møte er lovlig satt",
|
83 |
+
"sentence_language_code": "nb-NO",
|
84 |
+
"text": "Stortingets møte er lovlig satt",
|
85 |
+
"start_time": 302650,
|
86 |
+
"end_time": 306000,
|
87 |
+
"normsentence_text": "Stortingets møte er lovlig satt",
|
88 |
+
"transsentence_text": "Stortingets møte er lovleg sett",
|
89 |
+
"translated": 1,
|
90 |
+
"audio": {
|
91 |
+
"path": "audio/20170207-095506_302650_306000.wav",
|
92 |
+
"array": [
|
93 |
+
24,
|
94 |
+
25,
|
95 |
+
50,
|
96 |
+
(...)
|
97 |
+
],
|
98 |
+
"sampling_rate": 48000
|
99 |
+
}
|
100 |
+
}
|
101 |
+
```
|
102 |
|
103 |
### Version B
|
104 |
This verison does not contain the audio encoded in the dataset. Instead it has the audio files placed in sub-directories. There are currently both samples in clips_48k_wav and clips_16k_mp3. Only the base filename is referred in the dataset. Please not that there are both sentence-based audio clips as well at meeting-based audio clips. The dataset contains referrals to both, the latter referral has start and stop time as well.
|
105 |
|
106 |
+
One line in the metadata.json looks like this:
|
107 |
+
```json
|
108 |
+
i{
|
109 |
+
"meeting_date": "20170207",
|
110 |
+
"full_audio_file": "20170207-095506",
|
111 |
+
"proceedings_file": "20170207-095506.ref",
|
112 |
+
"duration": 4442474,
|
113 |
+
"transcriber_id": 1,
|
114 |
+
"reviewer_id": 2,
|
115 |
+
"data_split": "test",
|
116 |
+
"speaker_name": "Marit Nybakk",
|
117 |
+
"speaker_id": 1,
|
118 |
+
"sentence_id": 7309,
|
119 |
+
"sentence_language_code": "nb-NO",
|
120 |
+
"sentence_text": "Stortingets møte er lovlig satt",
|
121 |
+
"sentence_order": 0,
|
122 |
+
"audio_file": "20170207-095506_302650_306000",
|
123 |
+
"start_time": 302650,
|
124 |
+
"end_time": 306000,
|
125 |
+
"normsentence_text": "Stortingets møte er lovlig satt",
|
126 |
+
"transsentence_text": "Stortingets møte er lovleg sett",
|
127 |
+
"translated": 1
|
128 |
+
}
|
129 |
+
```
|
130 |
+
|
131 |
|
132 |
### Dataset Creation
|
133 |
We are providing a **train**, **dev** and **test** split. These are the same as in the orginal corpus.
|
x.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:950bb11d677996a46daa0eb6c6d2133791abc8e9894155b0cbd816d5b36fd4d2
|
3 |
+
size 582
|