Update multilingual-TEDX-fr.py
Browse files- multilingual-TEDX-fr.py +16 -5
multilingual-TEDX-fr.py
CHANGED
@@ -52,13 +52,14 @@ class TEDXConfig(datasets.BuilderConfig):
|
|
52 |
)
|
53 |
self.single_samples = (name == "single_samples")
|
54 |
self.max = (name == "max")
|
|
|
55 |
if not self.single_samples and not self.max:
|
56 |
self.max_duration = float(name.split("=")[1][:-1])
|
57 |
-
else:
|
58 |
-
self.max_duration = np.inf
|
59 |
-
|
60 |
|
61 |
class TEDX(datasets.GeneratorBasedBuilder):
|
|
|
|
|
|
|
62 |
|
63 |
BUILDER_CONFIGS = [
|
64 |
TEDXConfig(name="single_samples", description="all samples taken separately, can be very short and imprecise"),
|
@@ -70,6 +71,9 @@ class TEDX(datasets.GeneratorBasedBuilder):
|
|
70 |
TEDXConfig(name="max=10s", description="samples are merged in order to reach a max duration of 10 seconds"
|
71 |
"Does not remove single utterances that may exceed "
|
72 |
"the maximum duration"),
|
|
|
|
|
|
|
73 |
]
|
74 |
|
75 |
DEFAULT_CONFIG_NAME = "single_samples"
|
@@ -152,7 +156,14 @@ class TEDX(datasets.GeneratorBasedBuilder):
|
|
152 |
]
|
153 |
|
154 |
return splits
|
155 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
156 |
@staticmethod
|
157 |
def merge_utterances(utterance1: Utterance, utterance2: Utterance) -> Utterance:
|
158 |
assert(utterance1.speaker_id == utterance2.speaker_id)
|
@@ -181,7 +192,7 @@ class TEDX(datasets.GeneratorBasedBuilder):
|
|
181 |
yield merged_utterance
|
182 |
return
|
183 |
end_time = new_utterance.end_timestamp
|
184 |
-
if end_time - start_time > self.
|
185 |
yield merged_utterance
|
186 |
merged_utterance = new_utterance
|
187 |
start_time = merged_utterance.start_timestamp
|
|
|
52 |
)
|
53 |
self.single_samples = (name == "single_samples")
|
54 |
self.max = (name == "max")
|
55 |
+
self.random = (name == "random_merge")
|
56 |
if not self.single_samples and not self.max:
|
57 |
self.max_duration = float(name.split("=")[1][:-1])
|
|
|
|
|
|
|
58 |
|
59 |
class TEDX(datasets.GeneratorBasedBuilder):
|
60 |
+
|
61 |
+
random_max_durations = cycle([8, 4, 10, 5, 13, 23, 6, 19, 24, 7, 26, 27, 20, 14, 1, 25, 21, 22,
|
62 |
+
9, 12, 11, 2, 30, 15, 28, 17, 18, 29, 16, 3])
|
63 |
|
64 |
BUILDER_CONFIGS = [
|
65 |
TEDXConfig(name="single_samples", description="all samples taken separately, can be very short and imprecise"),
|
|
|
71 |
TEDXConfig(name="max=10s", description="samples are merged in order to reach a max duration of 10 seconds"
|
72 |
"Does not remove single utterances that may exceed "
|
73 |
"the maximum duration"),
|
74 |
+
TEDXConfig(name="random_merge", description="samples are merged in order to reach a random max duration between 1 and 30 seconds"
|
75 |
+
"Does not remove single utterances that may exceed "
|
76 |
+
"the maximum duration"),
|
77 |
]
|
78 |
|
79 |
DEFAULT_CONFIG_NAME = "single_samples"
|
|
|
156 |
]
|
157 |
|
158 |
return splits
|
159 |
+
|
160 |
+
def get_max_duration(self) -> float:
|
161 |
+
if self.config.max:
|
162 |
+
return np.inf
|
163 |
+
if self.config.random:
|
164 |
+
return next(self.random_max_durations)
|
165 |
+
return self.config.max_duration
|
166 |
+
|
167 |
@staticmethod
|
168 |
def merge_utterances(utterance1: Utterance, utterance2: Utterance) -> Utterance:
|
169 |
assert(utterance1.speaker_id == utterance2.speaker_id)
|
|
|
192 |
yield merged_utterance
|
193 |
return
|
194 |
end_time = new_utterance.end_timestamp
|
195 |
+
if end_time - start_time > self.get_max_duration():
|
196 |
yield merged_utterance
|
197 |
merged_utterance = new_utterance
|
198 |
start_time = merged_utterance.start_timestamp
|