Update libriheavy.py
Browse files- libriheavy.py +59 -15
libriheavy.py
CHANGED
@@ -54,6 +54,7 @@ class Libriheavy(datasets.GeneratorBasedBuilder):
|
|
54 |
{
|
55 |
"id": datasets.Value("string"),
|
56 |
"speaker_id": datasets.Value("string"),
|
|
|
57 |
"audio": datasets.Value("string"),
|
58 |
"text": datasets.Value("string"),
|
59 |
"word_segments": datasets.Sequence(
|
@@ -71,6 +72,22 @@ class Libriheavy(datasets.GeneratorBasedBuilder):
|
|
71 |
}
|
72 |
),
|
73 |
"mel_spectrogram": datasets.Sequence(datasets.Sequence(datasets.Value("float32"))),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
74 |
}
|
75 |
),
|
76 |
supervised_keys=None,
|
@@ -153,10 +170,11 @@ class Libriheavy(datasets.GeneratorBasedBuilder):
|
|
153 |
# skip the last utterance
|
154 |
if utterance_id == sorted(list(text.keys()))[-1]:
|
155 |
continue
|
156 |
-
|
157 |
result = {
|
158 |
"id": chunk["speaker_id"] + "_" + utterance_id,
|
159 |
"speaker_id": chunk["speaker_id"],
|
|
|
160 |
"audio": chunk["audio"],
|
161 |
"text": chunk["text"],
|
162 |
"word_segments": [
|
@@ -165,24 +183,50 @@ class Libriheavy(datasets.GeneratorBasedBuilder):
|
|
165 |
"phone_segments": [
|
166 |
{"start": segment[0], "end": segment[1], "phone": segment[2]} for segment in utterance["phone_segments"]
|
167 |
],
|
168 |
-
"mel_spectrogram":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
169 |
}
|
170 |
yield chunk["speaker_id"] + "_" + utterance_id, result
|
171 |
else:
|
172 |
# only use the last utterance
|
173 |
utterance_id = sorted(list(text.keys()))[-1]
|
174 |
utterance = text[utterance_id]
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
188 |
yield chunk["speaker_id"] + "_" + utterance_id, result
|
|
|
54 |
{
|
55 |
"id": datasets.Value("string"),
|
56 |
"speaker_id": datasets.Value("string"),
|
57 |
+
"speaker_vec": datasets.Sequence(datasets.Value("float32")),
|
58 |
"audio": datasets.Value("string"),
|
59 |
"text": datasets.Value("string"),
|
60 |
"word_segments": datasets.Sequence(
|
|
|
72 |
}
|
73 |
),
|
74 |
"mel_spectrogram": datasets.Sequence(datasets.Sequence(datasets.Value("float32"))),
|
75 |
+
"attributes": datasets.Sequence(
|
76 |
+
{
|
77 |
+
"pitch": datasets.Sequence(datasets.Value("float32")),
|
78 |
+
"energy": datasets.Sequence(datasets.Value("float32")),
|
79 |
+
"snr": datasets.Sequence(datasets.Value("float32")),
|
80 |
+
"srmr": datasets.Sequence(datasets.Value("float32")),
|
81 |
+
}
|
82 |
+
),
|
83 |
+
"overall_attributes": datasets.Sequence(
|
84 |
+
{
|
85 |
+
"pitch": datasets.Value("float32"),
|
86 |
+
"energy": datasets.Value("float32"),
|
87 |
+
"snr": datasets.Value("float32"),
|
88 |
+
"srmr": datasets.Value("float32"),
|
89 |
+
}
|
90 |
+
),
|
91 |
}
|
92 |
),
|
93 |
supervised_keys=None,
|
|
|
170 |
# skip the last utterance
|
171 |
if utterance_id == sorted(list(text.keys()))[-1]:
|
172 |
continue
|
173 |
+
npz_item = npz[str(utterance_id)].item()
|
174 |
result = {
|
175 |
"id": chunk["speaker_id"] + "_" + utterance_id,
|
176 |
"speaker_id": chunk["speaker_id"],
|
177 |
+
"speaker_vec": npz_item["d_vector"],
|
178 |
"audio": chunk["audio"],
|
179 |
"text": chunk["text"],
|
180 |
"word_segments": [
|
|
|
183 |
"phone_segments": [
|
184 |
{"start": segment[0], "end": segment[1], "phone": segment[2]} for segment in utterance["phone_segments"]
|
185 |
],
|
186 |
+
"mel_spectrogram": npz_item["mel"][0][0],
|
187 |
+
"attributes": {
|
188 |
+
npz_item["pitch"],
|
189 |
+
npz_item["energy"],
|
190 |
+
npz_item["snr"],
|
191 |
+
npz_item["srmr"],
|
192 |
+
},
|
193 |
+
"overall_attributes": {
|
194 |
+
npz_item["overall_pitch"],
|
195 |
+
npz_item["overall_energy"],
|
196 |
+
npz_item["overall_snr"],
|
197 |
+
npz_item["overall_srmr"],
|
198 |
+
},
|
199 |
}
|
200 |
yield chunk["speaker_id"] + "_" + utterance_id, result
|
201 |
else:
|
202 |
# only use the last utterance
|
203 |
utterance_id = sorted(list(text.keys()))[-1]
|
204 |
utterance = text[utterance_id]
|
205 |
+
npz_item = npz[str(utterance_id)].item()
|
206 |
+
result = {
|
207 |
+
"id": chunk["speaker_id"] + "_" + utterance_id,
|
208 |
+
"speaker_id": chunk["speaker_id"],
|
209 |
+
"speaker_vec": npz_item["d_vector"],
|
210 |
+
"audio": chunk["audio"],
|
211 |
+
"text": chunk["text"],
|
212 |
+
"word_segments": [
|
213 |
+
{"start": segment[0], "end": segment[1], "word": segment[2]} for segment in utterance["word_segments"]
|
214 |
+
],
|
215 |
+
"phone_segments": [
|
216 |
+
{"start": segment[0], "end": segment[1], "phone": segment[2]} for segment in utterance["phone_segments"]
|
217 |
+
],
|
218 |
+
"mel_spectrogram": npz_item["mel"][0][0],
|
219 |
+
"attributes": {
|
220 |
+
npz_item["pitch"],
|
221 |
+
npz_item["energy"],
|
222 |
+
npz_item["snr"],
|
223 |
+
npz_item["srmr"],
|
224 |
+
},
|
225 |
+
"overall_attributes": {
|
226 |
+
npz_item["overall_pitch"],
|
227 |
+
npz_item["overall_energy"],
|
228 |
+
npz_item["overall_snr"],
|
229 |
+
npz_item["overall_srmr"],
|
230 |
+
},
|
231 |
+
}
|
232 |
yield chunk["speaker_id"] + "_" + utterance_id, result
|