zyk21 commited on
Commit
98449bb
·
verified ·
1 Parent(s): 5b479d4

Add packaged DubBench media archives and documentation

Browse files
README.md CHANGED
@@ -55,15 +55,15 @@ The Chinese subset comes from **CMLR (Chinese Mandarin Lip Reading)**. See the [
55
  dubbench/
56
  |-- english/
57
  | |-- metadata.tsv # 978 target/reference pairs
58
- | |-- target/ # 978 target utterances
59
- | `-- reference/ # 923 unique reference utterances
60
  `-- chinese/
61
  |-- metadata.tsv # 1,000 target/reference pairs
62
- |-- target/ # 1,000 target utterances
63
- `-- reference/ # 315 unique reference utterances
64
  ```
65
 
66
- Utterances normally have three sidecar files with the same stem:
67
 
68
  - `.wav`: audio waveform
69
  - `.mp4`: video clip
@@ -82,7 +82,7 @@ The TSV files are UTF-8, have no header row, and contain exactly four tab-separa
82
  | 3 | `reference_id` | Extensionless paired reference/source-utterance path relative to the language subset directory. |
83
  | 4 | `reference_text` | Transcript paired with the reference/source utterance. |
84
 
85
- There are no reserved, empty, or audio-root columns in the supplied files. The language subset directory is the path root. Append `.wav`, `.mp4`, or `.txt` to an ID to locate the corresponding asset.
86
 
87
  Example row (spaces are shown around the tab separators for readability):
88
 
@@ -96,12 +96,15 @@ Download only the DubBench directory with the Hugging Face CLI:
96
 
97
  ```bash
98
  hf download zyk21/VoxDub --repo-type dataset --include "dubbench/**" --local-dir VoxDub
 
 
99
  ```
100
 
101
- Or download it programmatically and resolve each extensionless ID against its language directory:
102
 
103
  ```python
104
  import csv
 
105
  from pathlib import Path
106
  from huggingface_hub import snapshot_download
107
 
@@ -112,6 +115,10 @@ repo_root = Path(snapshot_download(
112
  ))
113
  subset_root = repo_root / "dubbench" / "english" # or "chinese"
114
 
 
 
 
 
115
  with (subset_root / "metadata.tsv").open(encoding="utf-8", newline="") as f:
116
  for target_id, target_text, reference_id, reference_text in csv.reader(f, delimiter="\t"):
117
  target_wav = subset_root / f"{target_id}.wav"
 
55
  dubbench/
56
  |-- english/
57
  | |-- metadata.tsv # 978 target/reference pairs
58
+ | |-- target.tar # extracts to target/ (978 utterances)
59
+ | `-- reference.tar # extracts to reference/ (923 utterances)
60
  `-- chinese/
61
  |-- metadata.tsv # 1,000 target/reference pairs
62
+ |-- target.tar # extracts to target/ (1,000 utterances)
63
+ `-- reference.tar # extracts to reference/ (315 utterances)
64
  ```
65
 
66
+ The four uncompressed TAR archives preserve the organized `target/` and `reference/` directories while keeping Hub commits reliable. After extraction, utterances normally have three sidecar files with the same stem:
67
 
68
  - `.wav`: audio waveform
69
  - `.mp4`: video clip
 
82
  | 3 | `reference_id` | Extensionless paired reference/source-utterance path relative to the language subset directory. |
83
  | 4 | `reference_text` | Transcript paired with the reference/source utterance. |
84
 
85
+ There are no reserved, empty, or audio-root columns in the supplied files. After extracting both archives for a language, the language subset directory is the path root. Append `.wav`, `.mp4`, or `.txt` to an ID to locate the corresponding asset.
86
 
87
  Example row (spaces are shown around the tab separators for readability):
88
 
 
96
 
97
  ```bash
98
  hf download zyk21/VoxDub --repo-type dataset --include "dubbench/**" --local-dir VoxDub
99
+ tar -xf VoxDub/dubbench/english/target.tar -C VoxDub/dubbench/english
100
+ tar -xf VoxDub/dubbench/english/reference.tar -C VoxDub/dubbench/english
101
  ```
102
 
103
+ Repeat the two extraction commands with `chinese` for the Chinese subset. Alternatively, download and extract a subset programmatically, then resolve each extensionless ID against its language directory:
104
 
105
  ```python
106
  import csv
107
+ import tarfile
108
  from pathlib import Path
109
  from huggingface_hub import snapshot_download
110
 
 
115
  ))
116
  subset_root = repo_root / "dubbench" / "english" # or "chinese"
117
 
118
+ for role in ("target", "reference"):
119
+ with tarfile.open(subset_root / f"{role}.tar") as archive:
120
+ archive.extractall(subset_root)
121
+
122
  with (subset_root / "metadata.tsv").open(encoding="utf-8", newline="") as f:
123
  for target_id, target_text, reference_id, reference_text in csv.reader(f, delimiter="\t"):
124
  target_wav = subset_root / f"{target_id}.wav"
dubbench/chinese/target.tar ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:09298c70113beb8c8969a68b59309618175a8b51b6cb50303924ddf434e198d8
3
+ size 1161406464
dubbench/english/reference.tar ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d8c4c5ca76109093a81058b4cc41515e806d2288741cbd932878f0c00c3c9409
3
+ size 1342859264
dubbench/english/target.tar ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca344398abc8a2bba485601fe767bdb8bce0d6613bf4b3e2afbbeeaab4e63c88
3
+ size 1606132224