martillopartbsc
commited on
Commit
•
7bcefc4
1
Parent(s):
0734d5a
change to process_text
Browse files- infer_onnx.py +8 -7
infer_onnx.py
CHANGED
@@ -24,8 +24,15 @@ def intersperse(lst, item):
|
|
24 |
return result
|
25 |
|
26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
def process_text(i: int, text: str, device: torch.device, cleaner:str):
|
28 |
print(f"[{i}] - Input text: {text}")
|
|
|
29 |
x = torch.tensor(
|
30 |
intersperse(text_to_sequence(text, [cleaner]), 0),
|
31 |
dtype=torch.long,
|
@@ -36,12 +43,6 @@ def process_text(i: int, text: str, device: torch.device, cleaner:str):
|
|
36 |
print(x_phones)
|
37 |
return x.numpy(), x_lengths.numpy()
|
38 |
|
39 |
-
def check_input(text):
|
40 |
-
if len(text) > 500:
|
41 |
-
raise gr.Error("Wrong input, more than 5 characters")
|
42 |
-
else:
|
43 |
-
return text
|
44 |
-
|
45 |
# paths
|
46 |
MODEL_PATH_MATCHA_MEL_BAL="matcha_multispeaker_cat_bal_opset_15_10_steps.onnx"
|
47 |
MODEL_PATH_MATCHA_MEL_CAT="matcha_multispeaker_cat_cen_opset_15_10_steps.onnx"
|
@@ -162,7 +163,7 @@ def tts(text:str, accent:str, spk_name:str, temperature:float, length_scale:floa
|
|
162 |
denoise=True
|
163 |
spk_id = speaker_id_dict[accent][spk_name]
|
164 |
sid = np.array([int(spk_id)]) if spk_id is not None else None
|
165 |
-
text_matcha , text_lengths = process_text(0,
|
166 |
model_matcha_mel = models[accent]
|
167 |
|
168 |
# MATCHA VOCOS
|
|
|
24 |
return result
|
25 |
|
26 |
|
27 |
+
def check_input(text):
|
28 |
+
if len(text) > 500:
|
29 |
+
raise gr.Error("Wrong input, more than 5 characters")
|
30 |
+
else:
|
31 |
+
return text
|
32 |
+
|
33 |
def process_text(i: int, text: str, device: torch.device, cleaner:str):
|
34 |
print(f"[{i}] - Input text: {text}")
|
35 |
+
check_input(text)
|
36 |
x = torch.tensor(
|
37 |
intersperse(text_to_sequence(text, [cleaner]), 0),
|
38 |
dtype=torch.long,
|
|
|
43 |
print(x_phones)
|
44 |
return x.numpy(), x_lengths.numpy()
|
45 |
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
# paths
|
47 |
MODEL_PATH_MATCHA_MEL_BAL="matcha_multispeaker_cat_bal_opset_15_10_steps.onnx"
|
48 |
MODEL_PATH_MATCHA_MEL_CAT="matcha_multispeaker_cat_cen_opset_15_10_steps.onnx"
|
|
|
163 |
denoise=True
|
164 |
spk_id = speaker_id_dict[accent][spk_name]
|
165 |
sid = np.array([int(spk_id)]) if spk_id is not None else None
|
166 |
+
text_matcha , text_lengths = process_text(0,text,"cpu",cleaner=cleaners[accent])
|
167 |
model_matcha_mel = models[accent]
|
168 |
|
169 |
# MATCHA VOCOS
|