StyleInstructCaps commited on
Commit
8b2bea8
·
verified ·
1 Parent(s): 563ce83

Add files using upload-large-folder tool

Browse files
README.md CHANGED
@@ -11,26 +11,26 @@ configs:
11
  path: data/StyleInstructCaps/train-*.parquet
12
  - split: eval
13
  path: data/StyleInstructCaps/eval-*.parquet
14
- - config_name: OODPrompts
15
  data_files:
16
  - split: test
17
- path: data/OODPrompts/test-*.parquet
18
- - config_name: MetaSet
19
  data_files:
20
  - split: test
21
- path: data/MetaSet/test-*.parquet
22
- - config_name: IDBalanced
23
  data_files:
24
  - split: test
25
- path: data/IDBalanced/test-*.parquet
26
- - config_name: UDVCTK
27
  data_files:
28
  - split: test
29
- path: data/UDVCTK/test-*.parquet
30
- - config_name: UnseenEval
31
  data_files:
32
  - split: test
33
- path: data/UnseenEval/test-*.parquet
34
  ---
35
 
36
  # StyleInstructCaps
@@ -41,18 +41,24 @@ Style-captioned speech metadata for speaking-style captioning and instruction-fo
41
 
42
  ## Overview
43
 
44
- This dataset release contains style captions, transcriptions, tags, and instruction prompts for speech captioning research. It includes:
45
 
46
- - **StyleInstructCaps** — large-scale training and evaluation splits with rich style annotations
47
- - **OODPrompts** — out-of-distribution prompt evaluation on VCTK (32 utterances, 36 instruction prompts per row)
48
- - **MetaSet** — evaluation subset derived from the Eval-500 benchmark (434 utterances with full style and instruction fields)
49
- - **IDBalanced** — in-distribution, accent/gender-balanced evaluation set
50
- - **UDVCTK** — VCTK evaluation with 10 speaker-idiosyncratic style prompts
51
- - **UnseenEval** — unseen GigaSpeech-style evaluation set with ground-truth captions
 
 
 
 
 
 
52
 
53
  This is a **metadata-only** release. Audio is **not** redistributed. Each row includes a `relative_audio_path` that should be resolved against the corresponding upstream audio corpus listed below.
54
 
55
- During release preparation, local absolute `audio_path` values are removed, mixed string/list fields are normalized to consistent types, and numeric fields (`duration`, `snr`, `utterance_pitch_mean`) are stored as `float64`.
56
 
57
  ## License
58
 
@@ -65,39 +71,46 @@ from datasets import load_dataset
65
 
66
  REPO = "<HF_ORG>/<DATASET_NAME>"
67
 
68
- # Load a specific config and split
69
  train = load_dataset(REPO, "StyleInstructCaps", split="train")
70
  eval_split = load_dataset(REPO, "StyleInstructCaps", split="eval")
 
 
 
 
 
71
 
72
- # Load evaluation configs
73
- ood = load_dataset(REPO, "OODPrompts", split="test")
74
- metaset = load_dataset(REPO, "MetaSet", split="test")
75
- id_balanced = load_dataset(REPO, "IDBalanced", split="test")
76
- ud_vctk = load_dataset(REPO, "UDVCTK", split="test")
77
- unseen = load_dataset(REPO, "UnseenEval", split="test")
78
-
79
- # Inspect one example
80
  print(train[0])
81
  ```
82
 
83
  ## Dataset Structure
84
 
85
- Each config is stored as parquet files under `data/<config>/`. Local absolute `audio_path` values from the original manifests are **not** included; use `relative_audio_path` to locate audio in upstream datasets.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86
 
87
  ### StyleInstructCaps
88
 
89
  | Split | Rows | Description |
90
  | ----- | ---- | ----------- |
91
- | `train` | 1,031,840 | Full-scale training set with style captions and tags (11 parquet shards) |
92
- | `eval` | 14,756 | Held-out evaluation set with style captions, tags, and instruction fields |
93
 
94
- **Shared columns (train and eval)**
95
 
96
  | Column | Type | Description |
97
  | ------ | ---- | ----------- |
98
- | `source` | string | Upstream dataset identifier (e.g., `emilia`, `voxceleb`, `ears`, `expresso`) |
99
- | `relative_audio_path` | string | Relative path to the audio file within the upstream corpus |
100
- | `transcription` | string | Transcript of the utterance |
101
  | `text_description` | list[string] | Natural-language style caption(s) |
102
  | `speaker_idiosyncratic_style` | string | Speaker identity style caption |
103
  | `situational_contextual_style` | string | Situational/contextual style caption |
@@ -105,24 +118,8 @@ Each config is stored as parquet files under `data/<config>/`. Local absolute `a
105
  | `linguistic_pragmatic_style` | string | Linguistic/pragmatic style caption |
106
  | `perceptual_listener_centric_style` | string | Listener-centric style caption |
107
  | `holistic_creative_synthesis` | string | Holistic creative synthesis caption |
108
- | `intrinsic_tags` | list[string] | Intrinsic style tags |
109
- | `situational_tags` | list[string] | Situational tags (empty list when absent) |
110
- | `basic_tags` | list[string] | Basic attribute tags |
111
- | `all_tags` | list[string] | Combined tag set |
112
- | `speakerid` | string | Speaker identifier |
113
- | `name` | string | Speaker name (when available) |
114
- | `duration` | float | Utterance duration in seconds |
115
- | `gender` | string | Gender label |
116
- | `accent` | string | Accent label (empty string when absent) |
117
- | `pitch` | string | Pitch descriptor |
118
- | `speaking_rate` | string | Speaking-rate descriptor |
119
- | `noise` | string | Recording environment descriptor |
120
- | `utterance_pitch_mean` | float | Mean pitch (Hz) |
121
- | `snr` | float | Signal-to-noise ratio |
122
- | `phonemes` | string | Phoneme sequence |
123
- | `task` | string | Task label |
124
-
125
- **Eval-only columns**
126
 
127
  | Column | Type | Description |
128
  | ------ | ---- | ----------- |
@@ -133,51 +130,51 @@ Each config is stored as parquet files under `data/<config>/`. Local absolute `a
133
  | `perceptual_listener_centric_style_instruction` | string | Instruction for listener-centric style |
134
  | `holistic_creative_synthesis_instruction` | string | Instruction for holistic synthesis |
135
 
136
- ### OODPrompts
137
 
138
  | Split | Rows | Description |
139
  | ----- | ---- | ----------- |
140
- | `test` | 32 | VCTK OOD prompt evaluation (36 instruction prompts per utterance) |
141
 
142
- **Core columns:** `source`, `relative_audio_path`, `speakerid`, `accent`, and `prompt1_speaker_idiosyncratic_style_instruction` through `prompt36_holistic_creative_synthesis_style_instruction`, covering speaker-idiosyncratic, situational, expressive, linguistic, perceptual, and holistic prompt families.
143
 
144
- ### MetaSet
145
 
146
  | Split | Rows | Description |
147
  | ----- | ---- | ----------- |
148
- | `test` | 434 | Eval-500 benchmark subset with full style annotations and instructions |
149
 
150
- Schema matches **StyleInstructCaps/eval** (all shared style caption fields plus the six `*_instruction` columns).
151
 
152
- ### IDBalanced
153
 
154
  | Split | Rows | Description |
155
  | ----- | ---- | ----------- |
156
- | `test` | 400 | Accent/gender-balanced in-distribution evaluation set |
157
 
158
- **Core columns:** `source`, `relative_audio_path`, `speakerid`, `accent`, `duration`, `snr`, and `prompt1_speaker_idiosyncratic_style_instruction` through `prompt10_speaker_idiosyncratic_style_instruction`.
159
 
160
- ### UDVCTK
161
 
162
  | Split | Rows | Description |
163
  | ----- | ---- | ----------- |
164
- | `test` | 320 | VCTK evaluation with 10 speaker-idiosyncratic prompts |
165
 
166
- **Core columns:** `source`, `relative_audio_path`, `speakerid`, `accent`, and `prompt1_speaker_idiosyncratic_style_instruction` through `prompt10_speaker_idiosyncratic_style_instruction`.
167
 
168
- ### UnseenEval
169
 
170
  | Split | Rows | Description |
171
  | ----- | ---- | ----------- |
172
- | `test` | 600 | Unseen GigaSpeech-style evaluation set |
173
 
174
- **Core columns**
175
 
176
  | Column | Type | Description |
177
  | ------ | ---- | ----------- |
178
  | `source` | string | Upstream dataset identifier (`eval_giga`) |
179
  | `relative_audio_path` | string | Relative path within the evaluation GigaSpeech subset |
180
- | `ground_truth_caption` | string | Reference style caption |
181
  | `holistic_creative_synthesis_instruction` | string | Instruction for holistic caption generation |
182
 
183
  ## Audio Sources
 
11
  path: data/StyleInstructCaps/train-*.parquet
12
  - split: eval
13
  path: data/StyleInstructCaps/eval-*.parquet
14
+ - config_name: StyleInstructCaps-MetaSet
15
  data_files:
16
  - split: test
17
+ path: data/StyleInstructCaps-MetaSet/test-*.parquet
18
+ - config_name: In_domain_balanced_set
19
  data_files:
20
  - split: test
21
+ path: data/In_domain_balanced_set/test-*.parquet
22
+ - config_name: VCTK_speaker_prompts
23
  data_files:
24
  - split: test
25
+ path: data/VCTK_speaker_prompts/test-*.parquet
26
+ - config_name: VCTK_unseen_prompts
27
  data_files:
28
  - split: test
29
+ path: data/VCTK_unseen_prompts/test-*.parquet
30
+ - config_name: Gigaspeech_SpeechCraft_captions
31
  data_files:
32
  - split: test
33
+ path: data/Gigaspeech_SpeechCraft_captions/test-*.parquet
34
  ---
35
 
36
  # StyleInstructCaps
 
41
 
42
  ## Overview
43
 
44
+ This dataset release contains speaking-style captions, caption-generation instructions (when available), and minimal utterance metadata for speech captioning research. Each config includes only:
45
 
46
+ - `source`, `relative_audio_path`, `transcription`, `speakerid`, `duration` (when present in the source manifest)
47
+ - Speaking-style caption fields (when present)
48
+ - Instruction fields for caption generation (when present)
49
+
50
+ Configs included:
51
+
52
+ - **StyleInstructCaps** — large-scale training and evaluation splits
53
+ - **StyleInstructCaps-MetaSet** — Eval-500 benchmark subset (434 utterances)
54
+ - **In-domain balanced set** — accent/gender-balanced in-distribution evaluation set
55
+ - **VCTK speaker prompts** — VCTK evaluation with 10 speaker-idiosyncratic style prompts
56
+ - **VCTK (unseen prompts)** — VCTK out-of-distribution prompt evaluation (32 utterances, 36 prompts per row)
57
+ - **Gigaspeech (with SpeechCraft captions)** — unseen GigaSpeech evaluation set with SpeechCraft reference captions
58
 
59
  This is a **metadata-only** release. Audio is **not** redistributed. Each row includes a `relative_audio_path` that should be resolved against the corresponding upstream audio corpus listed below.
60
 
61
+ During release preparation, local absolute `audio_path` values are removed, mixed string/list caption fields are normalized to consistent types, and `duration` is stored as `float64`.
62
 
63
  ## License
64
 
 
71
 
72
  REPO = "<HF_ORG>/<DATASET_NAME>"
73
 
 
74
  train = load_dataset(REPO, "StyleInstructCaps", split="train")
75
  eval_split = load_dataset(REPO, "StyleInstructCaps", split="eval")
76
+ metaset = load_dataset(REPO, "StyleInstructCaps-MetaSet", split="test")
77
+ in_domain = load_dataset(REPO, "In_domain_balanced_set", split="test")
78
+ vctk_speaker = load_dataset(REPO, "VCTK_speaker_prompts", split="test")
79
+ vctk_unseen = load_dataset(REPO, "VCTK_unseen_prompts", split="test")
80
+ gigaspeech = load_dataset(REPO, "Gigaspeech_SpeechCraft_captions", split="test")
81
 
 
 
 
 
 
 
 
 
82
  print(train[0])
83
  ```
84
 
85
  ## Dataset Structure
86
 
87
+ Each config is stored as parquet files under `data/<config>/`. Tags, phonemes, SNR, gender/accent labels, and other non-caption metadata from the source manifests are **not** included in this release.
88
+
89
+ ### Shared release columns
90
+
91
+ These columns appear in a config when present in the source manifest:
92
+
93
+ | Column | Type | Description |
94
+ | ------ | ---- | ----------- |
95
+ | `source` | string | Upstream dataset identifier |
96
+ | `relative_audio_path` | string | Relative path to the audio file within the upstream corpus |
97
+ | `transcription` | string | Transcript of the utterance |
98
+ | `speakerid` | string | Speaker identifier |
99
+ | `duration` | float | Utterance duration in seconds |
100
+
101
+ Style caption and instruction columns are config-specific; see below.
102
 
103
  ### StyleInstructCaps
104
 
105
  | Split | Rows | Description |
106
  | ----- | ---- | ----------- |
107
+ | `train` | 1,031,840 | Full-scale training set (11 parquet shards) |
108
+ | `eval` | 14,756 | Held-out evaluation set with instructions |
109
 
110
+ **Style caption columns (train and eval):**
111
 
112
  | Column | Type | Description |
113
  | ------ | ---- | ----------- |
 
 
 
114
  | `text_description` | list[string] | Natural-language style caption(s) |
115
  | `speaker_idiosyncratic_style` | string | Speaker identity style caption |
116
  | `situational_contextual_style` | string | Situational/contextual style caption |
 
118
  | `linguistic_pragmatic_style` | string | Linguistic/pragmatic style caption |
119
  | `perceptual_listener_centric_style` | string | Listener-centric style caption |
120
  | `holistic_creative_synthesis` | string | Holistic creative synthesis caption |
121
+
122
+ **Instruction columns (eval only):**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
123
 
124
  | Column | Type | Description |
125
  | ------ | ---- | ----------- |
 
130
  | `perceptual_listener_centric_style_instruction` | string | Instruction for listener-centric style |
131
  | `holistic_creative_synthesis_instruction` | string | Instruction for holistic synthesis |
132
 
133
+ ### StyleInstructCaps-MetaSet
134
 
135
  | Split | Rows | Description |
136
  | ----- | ---- | ----------- |
137
+ | `test` | 434 | Eval-500 benchmark subset |
138
 
139
+ Schema matches **StyleInstructCaps/eval** (base columns, all style caption fields, and all six `*_instruction` columns).
140
 
141
+ ### In-domain balanced set
142
 
143
  | Split | Rows | Description |
144
  | ----- | ---- | ----------- |
145
+ | `test` | 400 | Accent/gender-balanced in-distribution evaluation set |
146
 
147
+ **Columns:** base metadata (`source`, `relative_audio_path`, `speakerid`, `duration`) plus `prompt1_speaker_idiosyncratic_style_instruction` through `prompt10_speaker_idiosyncratic_style_instruction`.
148
 
149
+ ### VCTK speaker prompts
150
 
151
  | Split | Rows | Description |
152
  | ----- | ---- | ----------- |
153
+ | `test` | 320 | VCTK evaluation with 10 speaker-idiosyncratic style prompts |
154
 
155
+ **Columns:** base metadata (`source`, `relative_audio_path`, `speakerid`) plus `prompt1_speaker_idiosyncratic_style_instruction` through `prompt10_speaker_idiosyncratic_style_instruction`.
156
 
157
+ ### VCTK (unseen prompts)
158
 
159
  | Split | Rows | Description |
160
  | ----- | ---- | ----------- |
161
+ | `test` | 32 | VCTK OOD prompt evaluation (36 instruction prompts per utterance) |
162
 
163
+ **Columns:** base metadata (`source`, `relative_audio_path`, `speakerid`) plus `prompt1_speaker_idiosyncratic_style_instruction` through `prompt36_holistic_creative_synthesis_style_instruction`.
164
 
165
+ ### Gigaspeech (with SpeechCraft captions)
166
 
167
  | Split | Rows | Description |
168
  | ----- | ---- | ----------- |
169
+ | `test` | 600 | Unseen GigaSpeech evaluation set |
170
 
171
+ **Columns:**
172
 
173
  | Column | Type | Description |
174
  | ------ | ---- | ----------- |
175
  | `source` | string | Upstream dataset identifier (`eval_giga`) |
176
  | `relative_audio_path` | string | Relative path within the evaluation GigaSpeech subset |
177
+ | `ground_truth_caption` | string | SpeechCraft reference style caption |
178
  | `holistic_creative_synthesis_instruction` | string | Instruction for holistic caption generation |
179
 
180
  ## Audio Sources
data/Gigaspeech_SpeechCraft_captions/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0df07446eed592dac46f6d4f6be2fd503ad205d35e87edb8a15f8ddba370ff70
3
+ size 61270
data/In_domain_balanced_set/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7fa0e7f8e6c8949f3e96b36349522d9eaa9a82cb7ba38ee2a50e94b91b867ca3
3
+ size 29880
data/StyleInstructCaps-MetaSet/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9242c2c4fbad451d2ea488b01a9035478bfedcad15a7615a85d892828bf8a0bf
3
+ size 359180
data/StyleInstructCaps/eval-00000-of-00001.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:17afd13d2949a63b6acacc44442ee097ab7a74c25558fe6a945c24bb045c42ab
3
- size 13078194
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:706394e1e721593dae5e930005f277298783e6be135176265677e29b61393c2c
3
+ size 11134420
data/StyleInstructCaps/train-00000-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:223c3dea5a5ece9b3cd1c7aef6f05914f8d05eca01a836c3e3c28734c94e36ee
3
- size 86122391
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed15af6ce2130ffb96df1cc78b60d802cee64ad877b06fdc6bbf4ece4969ce02
3
+ size 70915714
data/StyleInstructCaps/train-00001-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b0a92b0d069c329d9d73719bca9d408010cc874ba4b0d0df085e3e058679d4d6
3
- size 86173266
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f462a1697cd83afab9d6a1550ce264102b8555585e0ed42a480b58f131939760
3
+ size 70937964
data/StyleInstructCaps/train-00002-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:829152601149795b140217a6d0afab9c4347d12172ee6d80989e1409dea5d0ae
3
- size 86277608
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:227b7890a7b29ebc3e0ebe7ff26dc46467d77b60aa242728d25c6e07a79a54b2
3
+ size 71013201
data/StyleInstructCaps/train-00003-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:779b61855a633524b87608b40cd70511b6dbc46948b1046ede3d8a9fd6175d37
3
- size 86168178
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:113cb0b459da5059594c770988a161e8139ab72538bccf882c1e7ee533a6b8a6
3
+ size 70935589
data/StyleInstructCaps/train-00004-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d027615e332bb0a50b8d3fe25fe5669ac5d3a49bf3ea549558cb48391bc2bdd9
3
- size 86306501
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7039dc904bdf7a76a2487904f49d8d1f9e6c95ff6f9bca39770a171f83715e07
3
+ size 71053586
data/StyleInstructCaps/train-00005-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:efdd068f79f3e1e5ce401598a7404480a1613e942b8c3d33400a4054557640c7
3
- size 86234437
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7020c2990b678c43234bd97107331c75e061320d891b03bf476fee59d39f751c
3
+ size 71001711
data/StyleInstructCaps/train-00006-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:01beff6135c363cfd0d8ee0ed1f006c69248b367d587d89020fc82292c33391a
3
- size 86234391
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42b436f9b2cc80c7da8f2fae0dec62ca1c8aeedd7af564dde400a7438c5d0415
3
+ size 71012499
data/StyleInstructCaps/train-00007-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d900689869b88e8c4845bbcc52533a71980c4763ed9df60b98a0ede98932c40e
3
- size 86278309
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19f3a9eec1189bc363cc1640af1b41a017e2a542d12bbc45fff9969e10986f33
3
+ size 71038322
data/StyleInstructCaps/train-00008-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d24f307fbfe2198075d8476e85753995ebe26b7ac4f4e3fff4c6b7eb98219c37
3
- size 86151181
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ee8b61cc318dc9e5d35928cd4cd3d78abf076d2ac1c5b60b5ec10816a52d8a85
3
+ size 70930255
data/StyleInstructCaps/train-00009-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e3953342fadd91f93ae732be6e6e223b0fcfacb24fba5f2071cb1718869308b9
3
- size 86057880
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:23633c1dd6bbdeaf4de1fd702992772930761abe7dfa272acb869c24abcc4e5a
3
+ size 72480944
data/StyleInstructCaps/train-00010-of-00011.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:55d1fd91917a1d13fd960fe3c447b432b2e37e6c2c1ea933b9a40d8ec26bfbf8
3
- size 27398529
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9591d310b66cc914e4c4bfd4e8865861032b2c8adb43f46162d5f769a658409f
3
+ size 23162000
data/VCTK_speaker_prompts/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b28c8b65e5ccffce3f1c92b5cbebafa9a3977aa742495a608a4f31253b34073c
3
+ size 23823
data/VCTK_unseen_prompts/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be0453b775b0a5cef02a51610dded8b8e980ba3070d343fd83d34247a45015a0
3
+ size 63223
scripts/prepare_release.py CHANGED
@@ -2,7 +2,7 @@
2
  """Convert source JSON manifests into Hugging Face release parquet files.
3
 
4
  Reads source files under data/ (read-only), drops local absolute audio_path,
5
- and writes parquet shards under release/data/<config>/.
6
  """
7
 
8
  from __future__ import annotations
@@ -23,60 +23,68 @@ DATA_ROOT = RELEASE_ROOT / "data"
23
 
24
  ROWS_PER_SHARD = 100_000
25
 
 
 
 
 
 
 
 
 
26
  # Expected row counts for verification.
27
  EXPECTED_COUNTS = {
28
- ("StyleInstructCaps", "train"): 1_031_840,
29
- ("StyleInstructCaps", "eval"): 14_756,
30
- ("OODPrompts", "test"): 32,
31
- ("MetaSet", "test"): 434,
32
- ("IDBalanced", "test"): 400,
33
- ("UDVCTK", "test"): 320,
34
- ("UnseenEval", "test"): 600,
35
  }
36
 
37
  RELEASE_SPECS: list[dict[str, Any]] = [
38
  {
39
- "config": "StyleInstructCaps",
40
  "split": "train",
41
  "source": PROJECT_ROOT / "data/data/train_fullscale_with_desc_v2.json",
42
  "stream": True,
43
  },
44
  {
45
- "config": "StyleInstructCaps",
46
  "split": "eval",
47
  "source": PROJECT_ROOT / "data/data/evaluation_holdout_with_desc_llama.json",
48
  "stream": False,
49
  },
50
  {
51
- "config": "OODPrompts",
52
  "split": "test",
53
  "source": PROJECT_ROOT
54
  / "data/evaluation/data_input_dir/vctk_manifest_2utt_36prompts_unseen.json",
55
  "stream": False,
56
  },
57
  {
58
- "config": "MetaSet",
59
  "split": "test",
60
  "source": PROJECT_ROOT
61
  / "data/evaluation/data_input_dir/tbd1_500_llama_global.json",
62
  "stream": False,
63
  },
64
  {
65
- "config": "IDBalanced",
66
  "split": "test",
67
  "source": PROJECT_ROOT
68
  / "data/evaluation/data_input_dir/balance_acccent_gender_eval_set2_with_fields.json",
69
  "stream": False,
70
  },
71
  {
72
- "config": "UDVCTK",
73
  "split": "test",
74
  "source": PROJECT_ROOT
75
  / "data/evaluation/data_input_dir/vctk_manifest_10prompts.json",
76
  "stream": False,
77
  },
78
  {
79
- "config": "UnseenEval",
80
  "split": "test",
81
  "source": PROJECT_ROOT
82
  / "data/evaluation/data_input_dir/giga_manifest.json",
@@ -84,24 +92,29 @@ RELEASE_SPECS: list[dict[str, Any]] = [
84
  },
85
  ]
86
 
 
 
 
 
 
 
 
87
 
88
- STYLE_TEXT_FIELDS = {
 
89
  "speaker_idiosyncratic_style",
90
  "situational_contextual_style",
91
  "expressive_emotional_style",
92
  "linguistic_pragmatic_style",
93
  "perceptual_listener_centric_style",
94
  "holistic_creative_synthesis",
 
95
  "captions",
96
  }
97
 
98
- LIST_FIELDS = {
99
- "text_description",
100
- "intrinsic_tags",
101
- "situational_tags",
102
- "basic_tags",
103
- "all_tags",
104
- }
105
 
106
 
107
  def _to_text(value: Any) -> str | None:
@@ -125,19 +138,22 @@ def _to_string_list(value: Any) -> list[str]:
125
  return [str(value)]
126
 
127
 
128
- FLOAT_FIELDS = {
129
- "duration",
130
- "snr",
131
- "utterance_pitch_mean",
132
- }
133
-
134
-
135
  def _to_float(value: Any) -> float | None:
136
  if value is None:
137
  return None
138
  return float(value)
139
 
140
 
 
 
 
 
 
 
 
 
 
 
141
  def normalize_record(record: dict[str, Any]) -> dict[str, Any]:
142
  """Normalize heterogeneous JSON values to parquet-friendly types."""
143
  cleaned = dict(record)
@@ -151,19 +167,20 @@ def normalize_record(record: dict[str, Any]) -> dict[str, Any]:
151
  if field in cleaned:
152
  cleaned[field] = _to_string_list(cleaned[field])
153
 
154
- if "accent" in cleaned and cleaned["accent"] is None:
155
- cleaned["accent"] = ""
156
-
157
- for field in FLOAT_FIELDS:
158
- if field in cleaned:
159
- cleaned[field] = _to_float(cleaned[field])
160
 
161
  return cleaned
162
 
163
 
 
 
 
 
 
164
  def sanitize_record(record: dict[str, Any]) -> dict[str, Any]:
165
- """Drop local absolute audio_path and normalize field types."""
166
- return normalize_record(record)
167
 
168
 
169
  def iter_annotation_records(source_path: Path) -> Iterator[dict[str, Any]]:
 
2
  """Convert source JSON manifests into Hugging Face release parquet files.
3
 
4
  Reads source files under data/ (read-only), drops local absolute audio_path,
5
+ filters to release columns only, and writes parquet shards under release/data/<config>/.
6
  """
7
 
8
  from __future__ import annotations
 
23
 
24
  ROWS_PER_SHARD = 100_000
25
 
26
+ # Hugging Face config names (directory names under release/data/).
27
+ CONFIG_STYLE_INSTRUCT_CAPS = "StyleInstructCaps"
28
+ CONFIG_STYLE_INSTRUCT_CAPS_METASET = "StyleInstructCaps-MetaSet"
29
+ CONFIG_IN_DOMAIN_BALANCED = "In_domain_balanced_set"
30
+ CONFIG_VCTK_SPEAKER_PROMPTS = "VCTK_speaker_prompts"
31
+ CONFIG_VCTK_UNSEEN_PROMPTS = "VCTK_unseen_prompts"
32
+ CONFIG_GIGASPEECH_SPEECHCRAFT = "Gigaspeech_SpeechCraft_captions"
33
+
34
  # Expected row counts for verification.
35
  EXPECTED_COUNTS = {
36
+ (CONFIG_STYLE_INSTRUCT_CAPS, "train"): 1_031_840,
37
+ (CONFIG_STYLE_INSTRUCT_CAPS, "eval"): 14_756,
38
+ (CONFIG_VCTK_UNSEEN_PROMPTS, "test"): 32,
39
+ (CONFIG_STYLE_INSTRUCT_CAPS_METASET, "test"): 434,
40
+ (CONFIG_IN_DOMAIN_BALANCED, "test"): 400,
41
+ (CONFIG_VCTK_SPEAKER_PROMPTS, "test"): 320,
42
+ (CONFIG_GIGASPEECH_SPEECHCRAFT, "test"): 600,
43
  }
44
 
45
  RELEASE_SPECS: list[dict[str, Any]] = [
46
  {
47
+ "config": CONFIG_STYLE_INSTRUCT_CAPS,
48
  "split": "train",
49
  "source": PROJECT_ROOT / "data/data/train_fullscale_with_desc_v2.json",
50
  "stream": True,
51
  },
52
  {
53
+ "config": CONFIG_STYLE_INSTRUCT_CAPS,
54
  "split": "eval",
55
  "source": PROJECT_ROOT / "data/data/evaluation_holdout_with_desc_llama.json",
56
  "stream": False,
57
  },
58
  {
59
+ "config": CONFIG_VCTK_UNSEEN_PROMPTS,
60
  "split": "test",
61
  "source": PROJECT_ROOT
62
  / "data/evaluation/data_input_dir/vctk_manifest_2utt_36prompts_unseen.json",
63
  "stream": False,
64
  },
65
  {
66
+ "config": CONFIG_STYLE_INSTRUCT_CAPS_METASET,
67
  "split": "test",
68
  "source": PROJECT_ROOT
69
  / "data/evaluation/data_input_dir/tbd1_500_llama_global.json",
70
  "stream": False,
71
  },
72
  {
73
+ "config": CONFIG_IN_DOMAIN_BALANCED,
74
  "split": "test",
75
  "source": PROJECT_ROOT
76
  / "data/evaluation/data_input_dir/balance_acccent_gender_eval_set2_with_fields.json",
77
  "stream": False,
78
  },
79
  {
80
+ "config": CONFIG_VCTK_SPEAKER_PROMPTS,
81
  "split": "test",
82
  "source": PROJECT_ROOT
83
  / "data/evaluation/data_input_dir/vctk_manifest_10prompts.json",
84
  "stream": False,
85
  },
86
  {
87
+ "config": CONFIG_GIGASPEECH_SPEECHCRAFT,
88
  "split": "test",
89
  "source": PROJECT_ROOT
90
  / "data/evaluation/data_input_dir/giga_manifest.json",
 
92
  },
93
  ]
94
 
95
+ BASE_FIELDS = {
96
+ "source",
97
+ "relative_audio_path",
98
+ "transcription",
99
+ "speakerid",
100
+ "duration",
101
+ }
102
 
103
+ STYLE_CAPTION_FIELDS = {
104
+ "text_description",
105
  "speaker_idiosyncratic_style",
106
  "situational_contextual_style",
107
  "expressive_emotional_style",
108
  "linguistic_pragmatic_style",
109
  "perceptual_listener_centric_style",
110
  "holistic_creative_synthesis",
111
+ "ground_truth_caption",
112
  "captions",
113
  }
114
 
115
+ STYLE_TEXT_FIELDS = STYLE_CAPTION_FIELDS - {"text_description"}
116
+
117
+ LIST_FIELDS = {"text_description"}
 
 
 
 
118
 
119
 
120
  def _to_text(value: Any) -> str | None:
 
138
  return [str(value)]
139
 
140
 
 
 
 
 
 
 
 
141
  def _to_float(value: Any) -> float | None:
142
  if value is None:
143
  return None
144
  return float(value)
145
 
146
 
147
+ def is_instruction_field(key: str) -> bool:
148
+ return key.endswith("_instruction") or (
149
+ key.startswith("prompt") and "_instruction" in key
150
+ )
151
+
152
+
153
+ def is_release_field(key: str) -> bool:
154
+ return key in BASE_FIELDS or key in STYLE_CAPTION_FIELDS or is_instruction_field(key)
155
+
156
+
157
  def normalize_record(record: dict[str, Any]) -> dict[str, Any]:
158
  """Normalize heterogeneous JSON values to parquet-friendly types."""
159
  cleaned = dict(record)
 
167
  if field in cleaned:
168
  cleaned[field] = _to_string_list(cleaned[field])
169
 
170
+ if "duration" in cleaned:
171
+ cleaned["duration"] = _to_float(cleaned["duration"])
 
 
 
 
172
 
173
  return cleaned
174
 
175
 
176
+ def filter_release_record(record: dict[str, Any]) -> dict[str, Any]:
177
+ """Keep only base metadata plus style captions and instructions."""
178
+ return {key: value for key, value in record.items() if is_release_field(key)}
179
+
180
+
181
  def sanitize_record(record: dict[str, Any]) -> dict[str, Any]:
182
+ """Normalize, drop audio_path, and filter to release columns."""
183
+ return filter_release_record(normalize_record(record))
184
 
185
 
186
  def iter_annotation_records(source_path: Path) -> Iterator[dict[str, Any]]: