init
Browse files- fetch_dataset_s2s.py +0 -1
- fetch_dataset_s2t.py +46 -83
- format_text.py +0 -1
- main_s2t.sh +1 -1
fetch_dataset_s2s.py
CHANGED
@@ -107,7 +107,6 @@ def cleanup(features, feature_file):
|
|
107 |
|
108 |
|
109 |
def get_audio(dataframe: pd.DataFrame):
|
110 |
-
resampler = {}
|
111 |
features = {"line_no": int(dataframe.pop('line_no').values[0])}
|
112 |
feature_file = p_join(cache_dir_feature, f'{features["line_no"]}.json')
|
113 |
for side, df in dataframe.groupby("side"):
|
|
|
107 |
|
108 |
|
109 |
def get_audio(dataframe: pd.DataFrame):
|
|
|
110 |
features = {"line_no": int(dataframe.pop('line_no').values[0])}
|
111 |
feature_file = p_join(cache_dir_feature, f'{features["line_no"]}.json')
|
112 |
for side, df in dataframe.groupby("side"):
|
fetch_dataset_s2t.py
CHANGED
@@ -19,18 +19,17 @@ from datasets import Dataset, Audio, DatasetDict
|
|
19 |
audio_loader = Audio()
|
20 |
# dataset config
|
21 |
url_metadata_dict = {
|
22 |
-
"enA-jaA": "https://dl.fbaipublicfiles.com/seamless/data/seamless_align_nov2023_extension/seamless.dataset.metadata.public.enA-jaA.tsv.gz",
|
23 |
"enA-jpn": "https://dl.fbaipublicfiles.com/seamless/data/seamless.dataset.metadata.public.enA-jpn.withduration.tsv.gz"
|
24 |
}
|
25 |
direction_speech = os.getenv("DIRECTION_SPEECH", "enA")
|
26 |
direction_text = os.getenv("DIRECTION_TEXT", "jpn")
|
27 |
-
direction =
|
28 |
-
|
29 |
-
|
30 |
cache_dir_feature = p_join("download", "feature", direction)
|
|
|
31 |
os.makedirs(cache_dir_feature, exist_ok=True)
|
32 |
-
|
33 |
-
os.makedirs(p_join(cache_dir_audio, s), exist_ok=True)
|
34 |
# processor config
|
35 |
n_pool = int(os.getenv("N_POOL", 1))
|
36 |
wget_max_retry = os.getenv("MAX_RETRY", "2")
|
@@ -43,6 +42,11 @@ hf_dataset = f"seamless-align-{direction}"
|
|
43 |
skip_download = bool(int(os.getenv("SKIP_DOWNLOAD", 0)))
|
44 |
sampling_rate = 16000 # seamless-align aligns audio in 16kHz
|
45 |
|
|
|
|
|
|
|
|
|
|
|
46 |
|
47 |
def wget(url: str, output_file: Optional[str] = None):
|
48 |
os.makedirs(os.path.dirname(output_file), exist_ok=True)
|
@@ -96,52 +100,49 @@ def to_json_serializable(val):
|
|
96 |
def cleanup(features, feature_file):
|
97 |
if os.path.exists(feature_file):
|
98 |
os.remove(feature_file)
|
99 |
-
for
|
100 |
-
|
101 |
-
os.remove(_unrelated_audio_file)
|
102 |
# create a dummy so that we can skip from next run
|
103 |
with open(feature_file, "w") as f:
|
104 |
json.dump({"dummy": "dummy"}, f)
|
105 |
|
106 |
|
107 |
def get_audio(dataframe: pd.DataFrame):
|
108 |
-
resampler = {}
|
109 |
features = {"line_no": int(dataframe.pop('line_no').values[0])}
|
|
|
|
|
|
|
110 |
feature_file = p_join(cache_dir_feature, f'{features["line_no"]}.json')
|
111 |
-
for
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
122 |
cleanup(features, feature_file)
|
123 |
return None
|
124 |
-
else:
|
125 |
-
try:
|
126 |
-
print(f"LOAD AUDIO FROM {features[f'{side}.path']}")
|
127 |
-
wav, sr = sf.read(features[f"{side}.path"])
|
128 |
-
print(f"wav shape:{wav.shape}")
|
129 |
-
if wav.ndim > 1:
|
130 |
-
wav = wav[:, 0]
|
131 |
-
wav = wav[floor(start / sampling_rate * sr):ceil(end / sampling_rate * sr)]
|
132 |
-
print(f"wav shape (after truncate):{wav.shape}")
|
133 |
-
wav = wav[:int(end/sampling_rate * sr) + sr]
|
134 |
-
print(f"SAVING: {features[f'{side}.path']}")
|
135 |
-
sf.write(features[f"{side}.path"], wav, sr)
|
136 |
-
# if sr != sampling_rate:
|
137 |
-
# print(f"RESAMPLING: {wav.shape} length audio")
|
138 |
-
# wav = librosa.resample(wav, orig_sr=sr, target_sr=sampling_rate)
|
139 |
-
# sf.write(features[f"{side}.path"], wav[start:end], sampling_rate)
|
140 |
-
|
141 |
-
except Exception as e:
|
142 |
-
print(f"\n#### ERROR ####\n {e}")
|
143 |
-
cleanup(features, feature_file)
|
144 |
-
return None
|
145 |
print(f"\n### SUCCESS! ###\n:{features['line_no']}")
|
146 |
with open(feature_file, "w") as f:
|
147 |
json.dump(features, f)
|
@@ -164,10 +165,7 @@ if __name__ == '__main__':
|
|
164 |
)
|
165 |
]
|
166 |
print(f"filtered unique lines: {len(inputs)}")
|
167 |
-
if
|
168 |
-
inputs = [g for g in inputs if len(g["side"].unique()) == 2 and set(g["side"].unique()) == sides]
|
169 |
-
print(f"removed side != 2: {len(inputs)}")
|
170 |
-
|
171 |
if n_pool == 1:
|
172 |
for g in tqdm(inputs, total=len(inputs)):
|
173 |
line_no = get_audio(g)
|
@@ -187,46 +185,11 @@ if __name__ == '__main__':
|
|
187 |
print(f"- dummy removed: {len(features)}")
|
188 |
print(f"push {len(features)} records to hub")
|
189 |
data_dict = {}
|
190 |
-
for
|
191 |
-
data_dict.update({f"{side}.audio": [i.pop(f"{side}.path") for i in features]})
|
192 |
data_dict.update({k: [i[k] for i in features] for k in features[0].keys()})
|
193 |
audio_dataset = Dataset.from_dict(data_dict)
|
194 |
-
|
195 |
-
audio_dataset = audio_dataset.cast_column(f"{side}.audio", Audio())
|
196 |
DatasetDict({"train": audio_dataset}).push_to_hub(
|
197 |
f"{hf_org}/{hf_dataset}",
|
198 |
config_name=f"subset_{dataset_id}"
|
199 |
)
|
200 |
-
|
201 |
-
|
202 |
-
# DatasetDict({"train": audio_dataset.select(list(range(1000)))}).push_to_hub(
|
203 |
-
# f"{hf_org}/{hf_dataset}",
|
204 |
-
# config_name=f"subset_{dataset_id}"
|
205 |
-
# )
|
206 |
-
|
207 |
-
# # 2 panel
|
208 |
-
# dataset_id = 75
|
209 |
-
# DatasetDict({"train": audio_dataset.select(list(range(3000, len(audio_dataset))))}).push_to_hub(
|
210 |
-
# f"{hf_org}/{hf_dataset}",
|
211 |
-
# config_name=f"subset_{dataset_id}"
|
212 |
-
# )
|
213 |
-
#
|
214 |
-
#
|
215 |
-
|
216 |
-
|
217 |
-
# audio_dataset = audio_dataset.select(list(range(2500)))
|
218 |
-
# dataset_to_push = DatasetDict({"train": audio_dataset})
|
219 |
-
# repo_name = f"{hf_org}/{hf_dataset}"
|
220 |
-
# dataset_to_push.push_to_hub(repo_name, config_name=f"subset_{dataset_id}")
|
221 |
-
# dataset_to_push.push_to_hub(repo_name, config_name=f"subset_{dataset_id}", max_shard_size="2GiB")
|
222 |
-
# dataset_to_push.push_to_hub(repo_name, config_name=f"subset_{dataset_id}", num_shards={"train": 1})
|
223 |
-
|
224 |
-
# while True:
|
225 |
-
# try:
|
226 |
-
# dataset_to_push.push_to_hub(repo_name, config_name=f"subset_{dataset_id}")
|
227 |
-
# break
|
228 |
-
# except Exception:
|
229 |
-
# print(f"FAILED: push_to_hub on {repo_name} failed. wait 60 sec and retry soon...")
|
230 |
-
# time.sleep(60)
|
231 |
-
|
232 |
-
|
|
|
19 |
audio_loader = Audio()
|
20 |
# dataset config
|
21 |
url_metadata_dict = {
|
|
|
22 |
"enA-jpn": "https://dl.fbaipublicfiles.com/seamless/data/seamless.dataset.metadata.public.enA-jpn.withduration.tsv.gz"
|
23 |
}
|
24 |
direction_speech = os.getenv("DIRECTION_SPEECH", "enA")
|
25 |
direction_text = os.getenv("DIRECTION_TEXT", "jpn")
|
26 |
+
direction = f"{direction_speech}-{direction_text}"
|
27 |
+
if direction not in url_metadata_dict:
|
28 |
+
url_metadata_dict[direction] = f"https://dl.fbaipublicfiles.com/seamless/data/seamless.dataset.metadata.public.{direction}.withduration.tsv.gz"
|
29 |
cache_dir_feature = p_join("download", "feature", direction)
|
30 |
+
cache_dir_audio = p_join("download", "audio", direction)
|
31 |
os.makedirs(cache_dir_feature, exist_ok=True)
|
32 |
+
os.makedirs(p_join(cache_dir_audio, direction_speech), exist_ok=True)
|
|
|
33 |
# processor config
|
34 |
n_pool = int(os.getenv("N_POOL", 1))
|
35 |
wget_max_retry = os.getenv("MAX_RETRY", "2")
|
|
|
42 |
skip_download = bool(int(os.getenv("SKIP_DOWNLOAD", 0)))
|
43 |
sampling_rate = 16000 # seamless-align aligns audio in 16kHz
|
44 |
|
45 |
+
text_corpus = p_join("text_corpus", f"text.{direction_speech}-{direction_text}.json")
|
46 |
+
assert os.path.exists(text_corpus)
|
47 |
+
with open(text_corpus) as f:
|
48 |
+
line_no_to_text = json.load(f)
|
49 |
+
|
50 |
|
51 |
def wget(url: str, output_file: Optional[str] = None):
|
52 |
os.makedirs(os.path.dirname(output_file), exist_ok=True)
|
|
|
100 |
def cleanup(features, feature_file):
|
101 |
if os.path.exists(feature_file):
|
102 |
os.remove(feature_file)
|
103 |
+
for _unrelated_audio_file in glob(p_join(cache_dir_audio, direction_speech, f"{features['line_no']}.*")):
|
104 |
+
os.remove(_unrelated_audio_file)
|
|
|
105 |
# create a dummy so that we can skip from next run
|
106 |
with open(feature_file, "w") as f:
|
107 |
json.dump({"dummy": "dummy"}, f)
|
108 |
|
109 |
|
110 |
def get_audio(dataframe: pd.DataFrame):
|
|
|
111 |
features = {"line_no": int(dataframe.pop('line_no').values[0])}
|
112 |
+
if features["line_no"] not in text_corpus:
|
113 |
+
return None
|
114 |
+
features[f"{direction_text}.text"] = text_corpus[features["line_no"]]
|
115 |
feature_file = p_join(cache_dir_feature, f'{features["line_no"]}.json')
|
116 |
+
features.update({f"{direction_speech}.{k}": to_json_serializable(v) for k, v in dataframe.iloc[0].to_dict().items()})
|
117 |
+
identifier = os.path.basename(features[f"{direction_speech}.url"]).split(".")[-1]
|
118 |
+
features[f"{direction_speech}.path"] = p_join(
|
119 |
+
cache_dir_audio, direction_speech, f"{features['line_no']}.{identifier}"
|
120 |
+
)
|
121 |
+
start, end = features[f"{direction_speech}.duration_start"], features[f"{direction_speech}.duration_end"]
|
122 |
+
if not os.path.exists(features[f"{direction_speech}.path"]):
|
123 |
+
print(f"WGET {features[f'{direction_speech}.url']}")
|
124 |
+
flag = wget(features[f"{direction_speech}.url"], output_file=features[f"{direction_speech}.path"])
|
125 |
+
if not flag:
|
126 |
+
print("\n#### ERROR: wget failure ####\n")
|
127 |
+
cleanup(features, feature_file)
|
128 |
+
return None
|
129 |
+
else:
|
130 |
+
try:
|
131 |
+
print(f"LOAD AUDIO FROM {features[f'{direction_speech}.path']}")
|
132 |
+
wav, sr = sf.read(features[f"{direction_speech}.path"])
|
133 |
+
print(f"wav shape:{wav.shape}")
|
134 |
+
if wav.ndim > 1:
|
135 |
+
wav = wav[:, 0]
|
136 |
+
wav = wav[floor(start / sampling_rate * sr):ceil(end / sampling_rate * sr)]
|
137 |
+
print(f"wav shape (after truncate):{wav.shape}")
|
138 |
+
wav = wav[:int(end/sampling_rate * sr) + sr]
|
139 |
+
print(f"SAVING: {features[f'{direction_speech}.path']}")
|
140 |
+
sf.write(features[f"{direction_speech}.path"], wav, sr)
|
141 |
+
|
142 |
+
except Exception as e:
|
143 |
+
print(f"\n#### ERROR ####\n {e}")
|
144 |
cleanup(features, feature_file)
|
145 |
return None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
146 |
print(f"\n### SUCCESS! ###\n:{features['line_no']}")
|
147 |
with open(feature_file, "w") as f:
|
148 |
json.dump(features, f)
|
|
|
165 |
)
|
166 |
]
|
167 |
print(f"filtered unique lines: {len(inputs)}")
|
168 |
+
inputs = [g for g in inputs if len(g) == 1]
|
|
|
|
|
|
|
169 |
if n_pool == 1:
|
170 |
for g in tqdm(inputs, total=len(inputs)):
|
171 |
line_no = get_audio(g)
|
|
|
185 |
print(f"- dummy removed: {len(features)}")
|
186 |
print(f"push {len(features)} records to hub")
|
187 |
data_dict = {}
|
188 |
+
data_dict.update({f"{direction_speech}.audio": [i.pop(f"{direction_speech}.path") for i in features]})
|
|
|
189 |
data_dict.update({k: [i[k] for i in features] for k in features[0].keys()})
|
190 |
audio_dataset = Dataset.from_dict(data_dict)
|
191 |
+
audio_dataset = audio_dataset.cast_column(f"{direction_speech}.audio", Audio())
|
|
|
192 |
DatasetDict({"train": audio_dataset}).push_to_hub(
|
193 |
f"{hf_org}/{hf_dataset}",
|
194 |
config_name=f"subset_{dataset_id}"
|
195 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
format_text.py
CHANGED
@@ -7,7 +7,6 @@ import pandas as pd
|
|
7 |
|
8 |
direction_speech = os.getenv("DIRECTION_SPEECH", "enA")
|
9 |
direction_text = os.getenv("DIRECTION_TEXT", "jpn")
|
10 |
-
direction = os.getenv("DIRECTION", "enA-jpn")
|
11 |
|
12 |
df = pd.concat([
|
13 |
pd.read_csv(i, quoting=csv.QUOTE_NONE, encoding='utf-8', sep='\t', header=None, on_bad_lines='skip')
|
|
|
7 |
|
8 |
direction_speech = os.getenv("DIRECTION_SPEECH", "enA")
|
9 |
direction_text = os.getenv("DIRECTION_TEXT", "jpn")
|
|
|
10 |
|
11 |
df = pd.concat([
|
12 |
pd.read_csv(i, quoting=csv.QUOTE_NONE, encoding='utf-8', sep='\t', header=None, on_bad_lines='skip')
|
main_s2t.sh
CHANGED
@@ -47,7 +47,7 @@ done
|
|
47 |
# text
|
48 |
export DIRECTION_SPEECH="enA"
|
49 |
export DIRECTION_TEXT="jpn"
|
50 |
-
export CHUNK_SIZE=
|
51 |
python download_s2t_metadata.py
|
52 |
for i in $(seq 1 ${CHUNK_SIZE});
|
53 |
do
|
|
|
47 |
# text
|
48 |
export DIRECTION_SPEECH="enA"
|
49 |
export DIRECTION_TEXT="jpn"
|
50 |
+
export CHUNK_SIZE=20
|
51 |
python download_s2t_metadata.py
|
52 |
for i in $(seq 1 ${CHUNK_SIZE});
|
53 |
do
|