init
Browse files- attach_speaker_embedding_s2s.py +0 -1
- main_s2s.sh +6 -4
- tokenize_dataset_s2s.py +2 -1
attach_speaker_embedding_s2s.py
CHANGED
@@ -60,7 +60,6 @@ print(f"Num examples (after filtering): {len(dataset)}")
|
|
60 |
|
61 |
def speaker_embedding(example):
|
62 |
for side in sides:
|
63 |
-
print(example[f"{side}.audio"]["array"].shape)
|
64 |
embedding = speaker_embedder.get_speaker_embedding(
|
65 |
example[f"{side}.audio"]["array"], example[f"{side}.audio"]["sampling_rate"]
|
66 |
)
|
|
|
60 |
|
61 |
def speaker_embedding(example):
|
62 |
for side in sides:
|
|
|
63 |
embedding = speaker_embedder.get_speaker_embedding(
|
64 |
example[f"{side}.audio"]["array"], example[f"{side}.audio"]["sampling_rate"]
|
65 |
)
|
main_s2s.sh
CHANGED
@@ -404,10 +404,12 @@ do
|
|
404 |
python attach_speaker_embedding_s2s.py
|
405 |
done
|
406 |
|
407 |
-
|
408 |
-
|
409 |
-
export
|
410 |
-
|
|
|
|
|
411 |
|
412 |
export SE_MODEL="w2vbert-600m" # running
|
413 |
export DATASET_ID=167
|
|
|
404 |
python attach_speaker_embedding_s2s.py
|
405 |
done
|
406 |
|
407 |
+
for i in 162 163 164 165 83 8;
|
408 |
+
do
|
409 |
+
export SE_MODEL="xlsr-2b" # running
|
410 |
+
export DIRECTION="enA-zhA"
|
411 |
+
python attach_speaker_embedding_s2s.py
|
412 |
+
done
|
413 |
|
414 |
export SE_MODEL="w2vbert-600m" # running
|
415 |
export DATASET_ID=167
|
tokenize_dataset_s2s.py
CHANGED
@@ -17,6 +17,7 @@ hf_dataset = f"seamless-align-{direction}"
|
|
17 |
dataset = load_dataset(f"{hf_org}/{hf_dataset}", f"subset_{dataset_id}", split="train")
|
18 |
tokenizer = EncodecTokenizer.from_pretrained()
|
19 |
max_seq_length = 10000000
|
|
|
20 |
audio_loader = Audio()
|
21 |
|
22 |
|
@@ -24,7 +25,7 @@ def error_file(example):
|
|
24 |
for side in sides:
|
25 |
try:
|
26 |
wav = audio_loader.decode_example(example[f"{side}.audio"])
|
27 |
-
if len(wav["array"])
|
28 |
return False
|
29 |
except ValueError:
|
30 |
return False
|
|
|
17 |
dataset = load_dataset(f"{hf_org}/{hf_dataset}", f"subset_{dataset_id}", split="train")
|
18 |
tokenizer = EncodecTokenizer.from_pretrained()
|
19 |
max_seq_length = 10000000
|
20 |
+
min_seq_length = 50000
|
21 |
audio_loader = Audio()
|
22 |
|
23 |
|
|
|
25 |
for side in sides:
|
26 |
try:
|
27 |
wav = audio_loader.decode_example(example[f"{side}.audio"])
|
28 |
+
if len(wav["array"]) < min_seq_length or len(wav["array"]) > max_seq_length:
|
29 |
return False
|
30 |
except ValueError:
|
31 |
return False
|