Spaces:
Running
Running
jhj0517
commited on
Commit
·
c8ead6d
1
Parent(s):
e76c01c
add base abstract class for translation model
Browse files- modules/translation_base.py +148 -0
modules/translation_base.py
ADDED
@@ -0,0 +1,148 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import torch
|
3 |
+
import gradio as gr
|
4 |
+
from abc import ABC, abstractmethod
|
5 |
+
from typing import List
|
6 |
+
from datetime import datetime
|
7 |
+
|
8 |
+
from modules.whisper_parameter import *
|
9 |
+
from modules.subtitle_manager import *
|
10 |
+
|
11 |
+
|
12 |
+
class TranslationBase(ABC):
|
13 |
+
def __init__(self,
|
14 |
+
model_dir: str):
|
15 |
+
super().__init__()
|
16 |
+
self.model = None
|
17 |
+
self.model_dir = model_dir
|
18 |
+
os.makedirs(self.model_dir, exist_ok=True)
|
19 |
+
self.current_model_size = None
|
20 |
+
self.device = self.get_device()
|
21 |
+
|
22 |
+
@abstractmethod
|
23 |
+
def translate(self,
|
24 |
+
text: str
|
25 |
+
):
|
26 |
+
pass
|
27 |
+
|
28 |
+
@abstractmethod
|
29 |
+
def update_model(self,
|
30 |
+
model_size: str,
|
31 |
+
src_lang: str,
|
32 |
+
tgt_lang: str,
|
33 |
+
progress: gr.Progress
|
34 |
+
):
|
35 |
+
pass
|
36 |
+
|
37 |
+
def translate_file(self,
|
38 |
+
fileobjs: list,
|
39 |
+
model_size: str,
|
40 |
+
src_lang: str,
|
41 |
+
tgt_lang: str,
|
42 |
+
add_timestamp: bool,
|
43 |
+
progress=gr.Progress()) -> list:
|
44 |
+
"""
|
45 |
+
Translate subtitle file from source language to target language
|
46 |
+
|
47 |
+
Parameters
|
48 |
+
----------
|
49 |
+
fileobjs: list
|
50 |
+
List of files to transcribe from gr.Files()
|
51 |
+
model_size: str
|
52 |
+
Whisper model size from gr.Dropdown()
|
53 |
+
src_lang: str
|
54 |
+
Source language of the file to translate from gr.Dropdown()
|
55 |
+
tgt_lang: str
|
56 |
+
Target language of the file to translate from gr.Dropdown()
|
57 |
+
add_timestamp: bool
|
58 |
+
Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the filename.
|
59 |
+
progress: gr.Progress
|
60 |
+
Indicator to show progress directly in gradio.
|
61 |
+
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
|
62 |
+
|
63 |
+
Returns
|
64 |
+
----------
|
65 |
+
A List of
|
66 |
+
String to return to gr.Textbox()
|
67 |
+
Files to return to gr.Files()
|
68 |
+
"""
|
69 |
+
try:
|
70 |
+
self.update_model(model_size=model_size,
|
71 |
+
src_lang=src_lang,
|
72 |
+
tgt_lang=tgt_lang,
|
73 |
+
progress=progress)
|
74 |
+
|
75 |
+
files_info = {}
|
76 |
+
for fileobj in fileobjs:
|
77 |
+
file_path = fileobj.name
|
78 |
+
file_name, file_ext = os.path.splitext(os.path.basename(fileobj.name))
|
79 |
+
if file_ext == ".srt":
|
80 |
+
parsed_dicts = parse_srt(file_path=file_path)
|
81 |
+
total_progress = len(parsed_dicts)
|
82 |
+
for index, dic in enumerate(parsed_dicts):
|
83 |
+
progress(index / total_progress, desc="Translating..")
|
84 |
+
translated_text = self.translate(dic["sentence"])
|
85 |
+
dic["sentence"] = translated_text
|
86 |
+
subtitle = get_serialized_srt(parsed_dicts)
|
87 |
+
|
88 |
+
timestamp = datetime.now().strftime("%m%d%H%M%S")
|
89 |
+
if add_timestamp:
|
90 |
+
output_path = os.path.join("outputs", "translations", f"{file_name}-{timestamp}")
|
91 |
+
else:
|
92 |
+
output_path = os.path.join("outputs", "translations", f"{file_name}.srt")
|
93 |
+
|
94 |
+
elif file_ext == ".vtt":
|
95 |
+
parsed_dicts = parse_vtt(file_path=file_path)
|
96 |
+
total_progress = len(parsed_dicts)
|
97 |
+
for index, dic in enumerate(parsed_dicts):
|
98 |
+
progress(index / total_progress, desc="Translating..")
|
99 |
+
translated_text = self.translate(dic["sentence"])
|
100 |
+
dic["sentence"] = translated_text
|
101 |
+
subtitle = get_serialized_vtt(parsed_dicts)
|
102 |
+
|
103 |
+
timestamp = datetime.now().strftime("%m%d%H%M%S")
|
104 |
+
if add_timestamp:
|
105 |
+
output_path = os.path.join("outputs", "translations", f"{file_name}-{timestamp}")
|
106 |
+
else:
|
107 |
+
output_path = os.path.join("outputs", "translations", f"{file_name}.vtt")
|
108 |
+
|
109 |
+
write_file(subtitle, output_path)
|
110 |
+
files_info[file_name] = subtitle
|
111 |
+
|
112 |
+
total_result = ''
|
113 |
+
for file_name, subtitle in files_info.items():
|
114 |
+
total_result += '------------------------------------\n'
|
115 |
+
total_result += f'{file_name}\n\n'
|
116 |
+
total_result += f'{subtitle}'
|
117 |
+
|
118 |
+
gr_str = f"Done! Subtitle is in the outputs/translation folder.\n\n{total_result}"
|
119 |
+
return [gr_str, output_path]
|
120 |
+
except Exception as e:
|
121 |
+
print(f"Error: {str(e)}")
|
122 |
+
finally:
|
123 |
+
self.release_cuda_memory()
|
124 |
+
self.remove_input_files([fileobj.name for fileobj in fileobjs])
|
125 |
+
|
126 |
+
@staticmethod
|
127 |
+
def get_device():
|
128 |
+
if torch.cuda.is_available():
|
129 |
+
return "cuda"
|
130 |
+
elif torch.backends.mps.is_available():
|
131 |
+
return "mps"
|
132 |
+
else:
|
133 |
+
return "cpu"
|
134 |
+
|
135 |
+
@staticmethod
|
136 |
+
def release_cuda_memory():
|
137 |
+
if torch.cuda.is_available():
|
138 |
+
torch.cuda.empty_cache()
|
139 |
+
torch.cuda.reset_max_memory_allocated()
|
140 |
+
|
141 |
+
@staticmethod
|
142 |
+
def remove_input_files(file_paths: List[str]):
|
143 |
+
if not file_paths:
|
144 |
+
return
|
145 |
+
|
146 |
+
for file_path in file_paths:
|
147 |
+
if file_path and os.path.exists(file_path):
|
148 |
+
os.remove(file_path)
|