Spaces:
Running
Running
jhj0517
commited on
Commit
·
9f69aa4
1
Parent(s):
7644f39
refactoring to use data class
Browse files- modules/whisper_Inference.py +142 -202
modules/whisper_Inference.py
CHANGED
@@ -10,6 +10,7 @@ import torch
|
|
10 |
from .base_interface import BaseInterface
|
11 |
from modules.subtitle_manager import get_srt, get_vtt, get_txt, write_file, safe_filename
|
12 |
from modules.youtube_manager import get_ytdata, get_ytaudio
|
|
|
13 |
|
14 |
DEFAULT_MODEL_SIZE = "large-v3"
|
15 |
|
@@ -21,82 +22,54 @@ class WhisperInference(BaseInterface):
|
|
21 |
self.model = None
|
22 |
self.available_models = whisper.available_models()
|
23 |
self.available_langs = sorted(list(whisper.tokenizer.LANGUAGES.values()))
|
|
|
24 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
25 |
self.available_compute_types = ["float16", "float32"]
|
26 |
self.current_compute_type = "float16" if self.device == "cuda" else "float32"
|
27 |
self.default_beam_size = 1
|
28 |
|
29 |
def transcribe_file(self,
|
30 |
-
|
31 |
-
model_size: str,
|
32 |
-
lang: str,
|
33 |
file_format: str,
|
34 |
-
istranslate: bool,
|
35 |
add_timestamp: bool,
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
compute_type: str,
|
40 |
-
progress=gr.Progress()) -> list:
|
41 |
"""
|
42 |
Write subtitle file from Files
|
43 |
|
44 |
Parameters
|
45 |
----------
|
46 |
-
|
47 |
List of files to transcribe from gr.Files()
|
48 |
-
model_size: str
|
49 |
-
Whisper model size from gr.Dropdown()
|
50 |
-
lang: str
|
51 |
-
Source language of the file to transcribe from gr.Dropdown()
|
52 |
file_format: str
|
53 |
-
File format to write from gr.Dropdown(). Supported format: [SRT, WebVTT, txt]
|
54 |
-
istranslate: bool
|
55 |
-
Boolean value from gr.Checkbox() that determines whether to translate to English.
|
56 |
-
It's Whisper's feature to translate speech from another language directly into English end-to-end.
|
57 |
add_timestamp: bool
|
58 |
-
Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the filename.
|
59 |
-
beam_size: int
|
60 |
-
Int value from gr.Number() that is used for decoding option.
|
61 |
-
log_prob_threshold: float
|
62 |
-
float value from gr.Number(). If the average log probability over sampled tokens is
|
63 |
-
below this value, treat as failed.
|
64 |
-
no_speech_threshold: float
|
65 |
-
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
66 |
-
the average log probability over sampled tokens is below `log_prob_threshold`,
|
67 |
-
consider the segment as silent.
|
68 |
-
compute_type: str
|
69 |
-
compute type from gr.Dropdown().
|
70 |
progress: gr.Progress
|
71 |
Indicator to show progress directly in gradio.
|
72 |
-
|
|
|
73 |
|
74 |
Returns
|
75 |
----------
|
76 |
-
|
77 |
-
|
78 |
-
|
|
|
79 |
"""
|
80 |
try:
|
81 |
-
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
|
82 |
-
|
83 |
files_info = {}
|
84 |
-
for
|
85 |
progress(0, desc="Loading Audio..")
|
86 |
-
audio = whisper.load_audio(
|
87 |
-
|
88 |
-
result, elapsed_time = self.transcribe(audio
|
89 |
-
|
90 |
-
|
91 |
-
beam_size=beam_size,
|
92 |
-
log_prob_threshold=log_prob_threshold,
|
93 |
-
no_speech_threshold=no_speech_threshold,
|
94 |
-
compute_type=compute_type,
|
95 |
-
progress=progress
|
96 |
-
)
|
97 |
progress(1, desc="Completed!")
|
98 |
|
99 |
-
file_name, file_ext = os.path.splitext(os.path.basename(
|
100 |
file_name = safe_filename(file_name)
|
101 |
subtitle, file_path = self.generate_and_write_file(
|
102 |
file_name=file_name,
|
@@ -104,7 +77,7 @@ class WhisperInference(BaseInterface):
|
|
104 |
add_timestamp=add_timestamp,
|
105 |
file_format=file_format
|
106 |
)
|
107 |
-
files_info[file_name] = {"subtitle": subtitle, "elapsed_time": elapsed_time, "path":
|
108 |
|
109 |
total_result = ''
|
110 |
total_time = 0
|
@@ -114,100 +87,71 @@ class WhisperInference(BaseInterface):
|
|
114 |
total_result += f"{info['subtitle']}"
|
115 |
total_time += info["elapsed_time"]
|
116 |
|
117 |
-
|
118 |
-
|
119 |
|
120 |
-
return [
|
121 |
except Exception as e:
|
122 |
print(f"Error transcribing file: {str(e)}")
|
123 |
finally:
|
124 |
self.release_cuda_memory()
|
125 |
-
self.remove_input_files([
|
126 |
|
127 |
def transcribe_youtube(self,
|
128 |
-
|
129 |
-
model_size: str,
|
130 |
-
lang: str,
|
131 |
file_format: str,
|
132 |
-
istranslate: bool,
|
133 |
add_timestamp: bool,
|
134 |
-
|
135 |
-
|
136 |
-
no_speech_threshold: float,
|
137 |
-
compute_type: str,
|
138 |
-
progress=gr.Progress()) -> list:
|
139 |
"""
|
140 |
Write subtitle file from Youtube
|
141 |
|
142 |
Parameters
|
143 |
----------
|
144 |
-
|
145 |
-
|
146 |
-
model_size: str
|
147 |
-
Whisper model size from gr.Dropdown()
|
148 |
-
lang: str
|
149 |
-
Source language of the file to transcribe from gr.Dropdown()
|
150 |
file_format: str
|
151 |
-
File format to write from gr.Dropdown(). Supported format: [SRT, WebVTT, txt]
|
152 |
-
istranslate: bool
|
153 |
-
Boolean value from gr.Checkbox() that determines whether to translate to English.
|
154 |
-
It's Whisper's feature to translate speech from another language directly into English end-to-end.
|
155 |
add_timestamp: bool
|
156 |
Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the filename.
|
157 |
-
beam_size: int
|
158 |
-
Int value from gr.Number() that is used for decoding option.
|
159 |
-
log_prob_threshold: float
|
160 |
-
float value from gr.Number(). If the average log probability over sampled tokens is
|
161 |
-
below this value, treat as failed.
|
162 |
-
no_speech_threshold: float
|
163 |
-
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
164 |
-
the average log probability over sampled tokens is below `log_prob_threshold`,
|
165 |
-
consider the segment as silent.
|
166 |
-
compute_type: str
|
167 |
-
compute type from gr.Dropdown().
|
168 |
progress: gr.Progress
|
169 |
Indicator to show progress directly in gradio.
|
170 |
-
|
|
|
171 |
|
172 |
Returns
|
173 |
----------
|
174 |
-
|
175 |
-
|
176 |
-
|
|
|
177 |
"""
|
178 |
try:
|
179 |
-
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
|
180 |
-
|
181 |
progress(0, desc="Loading Audio from Youtube..")
|
182 |
-
yt = get_ytdata(
|
183 |
audio = whisper.load_audio(get_ytaudio(yt))
|
184 |
|
185 |
-
result, elapsed_time = self.transcribe(audio
|
186 |
-
|
187 |
-
|
188 |
-
beam_size=beam_size,
|
189 |
-
log_prob_threshold=log_prob_threshold,
|
190 |
-
no_speech_threshold=no_speech_threshold,
|
191 |
-
compute_type=compute_type,
|
192 |
-
progress=progress)
|
193 |
progress(1, desc="Completed!")
|
194 |
|
195 |
file_name = safe_filename(yt.title)
|
196 |
-
subtitle,
|
197 |
file_name=file_name,
|
198 |
transcribed_segments=result,
|
199 |
add_timestamp=add_timestamp,
|
200 |
file_format=file_format
|
201 |
)
|
202 |
|
203 |
-
|
204 |
-
return [
|
205 |
except Exception as e:
|
206 |
print(f"Error transcribing youtube video: {str(e)}")
|
207 |
finally:
|
208 |
try:
|
209 |
if 'yt' not in locals():
|
210 |
-
yt = get_ytdata(
|
211 |
file_path = get_ytaudio(yt)
|
212 |
else:
|
213 |
file_path = get_ytaudio(yt)
|
@@ -218,116 +162,71 @@ class WhisperInference(BaseInterface):
|
|
218 |
pass
|
219 |
|
220 |
def transcribe_mic(self,
|
221 |
-
|
222 |
-
model_size: str,
|
223 |
-
lang: str,
|
224 |
file_format: str,
|
225 |
-
|
226 |
-
|
227 |
-
log_prob_threshold: float,
|
228 |
-
no_speech_threshold: float,
|
229 |
-
compute_type: str,
|
230 |
-
progress=gr.Progress()) -> list:
|
231 |
"""
|
232 |
Write subtitle file from microphone
|
233 |
|
234 |
Parameters
|
235 |
----------
|
236 |
-
|
237 |
Audio file path from gr.Microphone()
|
238 |
-
model_size: str
|
239 |
-
Whisper model size from gr.Dropdown()
|
240 |
-
lang: str
|
241 |
-
Source language of the file to transcribe from gr.Dropdown()
|
242 |
file_format: str
|
243 |
-
Subtitle format to write from gr.Dropdown(). Supported format: [SRT, WebVTT, txt]
|
244 |
-
istranslate: bool
|
245 |
-
Boolean value from gr.Checkbox() that determines whether to translate to English.
|
246 |
-
It's Whisper's feature to translate speech from another language directly into English end-to-end.
|
247 |
-
beam_size: int
|
248 |
-
Int value from gr.Number() that is used for decoding option.
|
249 |
-
log_prob_threshold: float
|
250 |
-
float value from gr.Number(). If the average log probability over sampled tokens is
|
251 |
-
below this value, treat as failed.
|
252 |
-
no_speech_threshold: float
|
253 |
-
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
254 |
-
the average log probability over sampled tokens is below `log_prob_threshold`,
|
255 |
-
consider the segment as silent.
|
256 |
-
compute_type: str
|
257 |
-
compute type from gr.Dropdown().
|
258 |
progress: gr.Progress
|
259 |
Indicator to show progress directly in gradio.
|
260 |
-
|
|
|
261 |
|
262 |
Returns
|
263 |
----------
|
264 |
-
|
265 |
-
|
266 |
-
|
|
|
267 |
"""
|
268 |
try:
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
log_prob_threshold=log_prob_threshold,
|
276 |
-
no_speech_threshold=no_speech_threshold,
|
277 |
-
compute_type=compute_type,
|
278 |
-
progress=progress)
|
279 |
progress(1, desc="Completed!")
|
280 |
|
281 |
-
subtitle,
|
282 |
file_name="Mic",
|
283 |
transcribed_segments=result,
|
284 |
add_timestamp=True,
|
285 |
file_format=file_format
|
286 |
)
|
287 |
|
288 |
-
|
289 |
-
return [
|
290 |
except Exception as e:
|
291 |
print(f"Error transcribing mic: {str(e)}")
|
292 |
finally:
|
293 |
self.release_cuda_memory()
|
294 |
-
self.remove_input_files([
|
295 |
|
296 |
def transcribe(self,
|
297 |
audio: Union[str, np.ndarray, torch.Tensor],
|
298 |
-
|
299 |
-
|
300 |
-
beam_size: int,
|
301 |
-
log_prob_threshold: float,
|
302 |
-
no_speech_threshold: float,
|
303 |
-
compute_type: str,
|
304 |
-
progress: gr.Progress
|
305 |
) -> Tuple[List[dict], float]:
|
306 |
"""
|
307 |
-
transcribe method for
|
308 |
|
309 |
Parameters
|
310 |
----------
|
311 |
-
audio: Union[str, BinaryIO,
|
312 |
Audio path or file binary or Audio numpy array
|
313 |
-
lang: str
|
314 |
-
Source language of the file to transcribe from gr.Dropdown()
|
315 |
-
istranslate: bool
|
316 |
-
Boolean value from gr.Checkbox() that determines whether to translate to English.
|
317 |
-
It's Whisper's feature to translate speech from another language directly into English end-to-end.
|
318 |
-
beam_size: int
|
319 |
-
Int value from gr.Number() that is used for decoding option.
|
320 |
-
log_prob_threshold: float
|
321 |
-
float value from gr.Number(). If the average log probability over sampled tokens is
|
322 |
-
below this value, treat as failed.
|
323 |
-
no_speech_threshold: float
|
324 |
-
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
325 |
-
the average log probability over sampled tokens is below `log_prob_threshold`,
|
326 |
-
consider the segment as silent.
|
327 |
-
compute_type: str
|
328 |
-
compute type from gr.Dropdown().
|
329 |
progress: gr.Progress
|
330 |
Indicator to show progress directly in gradio.
|
|
|
|
|
331 |
|
332 |
Returns
|
333 |
----------
|
@@ -337,45 +236,56 @@ class WhisperInference(BaseInterface):
|
|
337 |
elapsed time for transcription
|
338 |
"""
|
339 |
start_time = time.time()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
340 |
|
341 |
def progress_callback(progress_value):
|
342 |
progress(progress_value, desc="Transcribing..")
|
343 |
|
344 |
-
if lang == "Automatic Detection":
|
345 |
-
lang = None
|
346 |
-
|
347 |
-
translatable_model = ["large", "large-v1", "large-v2", "large-v3"]
|
348 |
segments_result = self.model.transcribe(audio=audio,
|
349 |
-
language=lang,
|
350 |
verbose=False,
|
351 |
-
beam_size=beam_size,
|
352 |
-
logprob_threshold=log_prob_threshold,
|
353 |
-
no_speech_threshold=no_speech_threshold,
|
354 |
-
task="translate" if
|
355 |
-
fp16=True if compute_type == "float16" else False,
|
356 |
progress_callback=progress_callback)["segments"]
|
357 |
elapsed_time = time.time() - start_time
|
358 |
|
359 |
return segments_result, elapsed_time
|
360 |
|
361 |
-
def
|
362 |
-
|
363 |
-
|
364 |
-
|
365 |
-
|
366 |
"""
|
367 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
368 |
"""
|
369 |
-
|
370 |
-
|
371 |
-
|
372 |
-
|
373 |
-
|
374 |
-
self.
|
375 |
-
|
376 |
-
|
377 |
-
download_root=os.path.join("models", "Whisper")
|
378 |
-
)
|
379 |
|
380 |
@staticmethod
|
381 |
def generate_and_write_file(file_name: str,
|
@@ -384,7 +294,25 @@ class WhisperInference(BaseInterface):
|
|
384 |
file_format: str,
|
385 |
) -> str:
|
386 |
"""
|
387 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
388 |
"""
|
389 |
timestamp = datetime.now().strftime("%m%d%H%M%S")
|
390 |
if add_timestamp:
|
@@ -410,6 +338,18 @@ class WhisperInference(BaseInterface):
|
|
410 |
|
411 |
@staticmethod
|
412 |
def format_time(elapsed_time: float) -> str:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
413 |
hours, rem = divmod(elapsed_time, 3600)
|
414 |
minutes, seconds = divmod(rem, 60)
|
415 |
|
|
|
10 |
from .base_interface import BaseInterface
|
11 |
from modules.subtitle_manager import get_srt, get_vtt, get_txt, write_file, safe_filename
|
12 |
from modules.youtube_manager import get_ytdata, get_ytaudio
|
13 |
+
from modules.whisper_data_class import *
|
14 |
|
15 |
DEFAULT_MODEL_SIZE = "large-v3"
|
16 |
|
|
|
22 |
self.model = None
|
23 |
self.available_models = whisper.available_models()
|
24 |
self.available_langs = sorted(list(whisper.tokenizer.LANGUAGES.values()))
|
25 |
+
self.translatable_model = ["large", "large-v1", "large-v2", "large-v3"]
|
26 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
27 |
self.available_compute_types = ["float16", "float32"]
|
28 |
self.current_compute_type = "float16" if self.device == "cuda" else "float32"
|
29 |
self.default_beam_size = 1
|
30 |
|
31 |
def transcribe_file(self,
|
32 |
+
files: list,
|
|
|
|
|
33 |
file_format: str,
|
|
|
34 |
add_timestamp: bool,
|
35 |
+
progress=gr.Progress(),
|
36 |
+
*whisper_params
|
37 |
+
) -> list:
|
|
|
|
|
38 |
"""
|
39 |
Write subtitle file from Files
|
40 |
|
41 |
Parameters
|
42 |
----------
|
43 |
+
files: list
|
44 |
List of files to transcribe from gr.Files()
|
|
|
|
|
|
|
|
|
45 |
file_format: str
|
46 |
+
Subtitle File format to write from gr.Dropdown(). Supported format: [SRT, WebVTT, txt]
|
|
|
|
|
|
|
47 |
add_timestamp: bool
|
48 |
+
Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the subtitle filename.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
progress: gr.Progress
|
50 |
Indicator to show progress directly in gradio.
|
51 |
+
*whisper_params: tuple
|
52 |
+
Gradio components related to Whisper. see whisper_data_class.py for details.
|
53 |
|
54 |
Returns
|
55 |
----------
|
56 |
+
result_str:
|
57 |
+
Result of transcription to return to gr.Textbox()
|
58 |
+
result_file_path:
|
59 |
+
Output file path to return to gr.Files()
|
60 |
"""
|
61 |
try:
|
|
|
|
|
62 |
files_info = {}
|
63 |
+
for file in files:
|
64 |
progress(0, desc="Loading Audio..")
|
65 |
+
audio = whisper.load_audio(file.name)
|
66 |
+
|
67 |
+
result, elapsed_time = self.transcribe(audio,
|
68 |
+
progress,
|
69 |
+
*whisper_params)
|
|
|
|
|
|
|
|
|
|
|
|
|
70 |
progress(1, desc="Completed!")
|
71 |
|
72 |
+
file_name, file_ext = os.path.splitext(os.path.basename(file.name))
|
73 |
file_name = safe_filename(file_name)
|
74 |
subtitle, file_path = self.generate_and_write_file(
|
75 |
file_name=file_name,
|
|
|
77 |
add_timestamp=add_timestamp,
|
78 |
file_format=file_format
|
79 |
)
|
80 |
+
files_info[file_name] = {"subtitle": subtitle, "elapsed_time": elapsed_time, "path": file_path}
|
81 |
|
82 |
total_result = ''
|
83 |
total_time = 0
|
|
|
87 |
total_result += f"{info['subtitle']}"
|
88 |
total_time += info["elapsed_time"]
|
89 |
|
90 |
+
result_str = f"Done in {self.format_time(total_time)}! Subtitle is in the outputs folder.\n\n{total_result}"
|
91 |
+
result_file_path = [info['path'] for info in files_info.values()]
|
92 |
|
93 |
+
return [result_str, result_file_path]
|
94 |
except Exception as e:
|
95 |
print(f"Error transcribing file: {str(e)}")
|
96 |
finally:
|
97 |
self.release_cuda_memory()
|
98 |
+
self.remove_input_files([file.name for file in files])
|
99 |
|
100 |
def transcribe_youtube(self,
|
101 |
+
youtube_link: str,
|
|
|
|
|
102 |
file_format: str,
|
|
|
103 |
add_timestamp: bool,
|
104 |
+
progress=gr.Progress(),
|
105 |
+
*whisper_params) -> list:
|
|
|
|
|
|
|
106 |
"""
|
107 |
Write subtitle file from Youtube
|
108 |
|
109 |
Parameters
|
110 |
----------
|
111 |
+
youtube_link: str
|
112 |
+
URL of the Youtube video to transcribe from gr.Textbox()
|
|
|
|
|
|
|
|
|
113 |
file_format: str
|
114 |
+
Subtitle File format to write from gr.Dropdown(). Supported format: [SRT, WebVTT, txt]
|
|
|
|
|
|
|
115 |
add_timestamp: bool
|
116 |
Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the filename.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
progress: gr.Progress
|
118 |
Indicator to show progress directly in gradio.
|
119 |
+
*whisper_params: tuple
|
120 |
+
Gradio components related to Whisper. see whisper_data_class.py for details.
|
121 |
|
122 |
Returns
|
123 |
----------
|
124 |
+
result_str:
|
125 |
+
Result of transcription to return to gr.Textbox()
|
126 |
+
result_file_path:
|
127 |
+
Output file path to return to gr.Files()
|
128 |
"""
|
129 |
try:
|
|
|
|
|
130 |
progress(0, desc="Loading Audio from Youtube..")
|
131 |
+
yt = get_ytdata(youtube_link)
|
132 |
audio = whisper.load_audio(get_ytaudio(yt))
|
133 |
|
134 |
+
result, elapsed_time = self.transcribe(audio,
|
135 |
+
progress,
|
136 |
+
*whisper_params)
|
|
|
|
|
|
|
|
|
|
|
137 |
progress(1, desc="Completed!")
|
138 |
|
139 |
file_name = safe_filename(yt.title)
|
140 |
+
subtitle, result_file_path = self.generate_and_write_file(
|
141 |
file_name=file_name,
|
142 |
transcribed_segments=result,
|
143 |
add_timestamp=add_timestamp,
|
144 |
file_format=file_format
|
145 |
)
|
146 |
|
147 |
+
result_str = f"Done in {self.format_time(elapsed_time)}! Subtitle file is in the outputs folder.\n\n{subtitle}"
|
148 |
+
return [result_str, result_file_path]
|
149 |
except Exception as e:
|
150 |
print(f"Error transcribing youtube video: {str(e)}")
|
151 |
finally:
|
152 |
try:
|
153 |
if 'yt' not in locals():
|
154 |
+
yt = get_ytdata(youtube_link)
|
155 |
file_path = get_ytaudio(yt)
|
156 |
else:
|
157 |
file_path = get_ytaudio(yt)
|
|
|
162 |
pass
|
163 |
|
164 |
def transcribe_mic(self,
|
165 |
+
mic_audio: str,
|
|
|
|
|
166 |
file_format: str,
|
167 |
+
progress=gr.Progress(),
|
168 |
+
*whisper_params) -> list:
|
|
|
|
|
|
|
|
|
169 |
"""
|
170 |
Write subtitle file from microphone
|
171 |
|
172 |
Parameters
|
173 |
----------
|
174 |
+
mic_audio: str
|
175 |
Audio file path from gr.Microphone()
|
|
|
|
|
|
|
|
|
176 |
file_format: str
|
177 |
+
Subtitle File format to write from gr.Dropdown(). Supported format: [SRT, WebVTT, txt]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
178 |
progress: gr.Progress
|
179 |
Indicator to show progress directly in gradio.
|
180 |
+
*whisper_params: tuple
|
181 |
+
Gradio components related to Whisper. see whisper_data_class.py for details.
|
182 |
|
183 |
Returns
|
184 |
----------
|
185 |
+
result_str:
|
186 |
+
Result of transcription to return to gr.Textbox()
|
187 |
+
result_file_path:
|
188 |
+
Output file path to return to gr.Files()
|
189 |
"""
|
190 |
try:
|
191 |
+
progress(0, desc="Loading Audio..")
|
192 |
+
result, elapsed_time = self.transcribe(
|
193 |
+
mic_audio,
|
194 |
+
progress,
|
195 |
+
*whisper_params,
|
196 |
+
)
|
|
|
|
|
|
|
|
|
197 |
progress(1, desc="Completed!")
|
198 |
|
199 |
+
subtitle, result_file_path = self.generate_and_write_file(
|
200 |
file_name="Mic",
|
201 |
transcribed_segments=result,
|
202 |
add_timestamp=True,
|
203 |
file_format=file_format
|
204 |
)
|
205 |
|
206 |
+
result_str = f"Done in {self.format_time(elapsed_time)}! Subtitle file is in the outputs folder.\n\n{subtitle}"
|
207 |
+
return [result_str, result_file_path]
|
208 |
except Exception as e:
|
209 |
print(f"Error transcribing mic: {str(e)}")
|
210 |
finally:
|
211 |
self.release_cuda_memory()
|
212 |
+
self.remove_input_files([mic_audio])
|
213 |
|
214 |
def transcribe(self,
|
215 |
audio: Union[str, np.ndarray, torch.Tensor],
|
216 |
+
progress: gr.Progress,
|
217 |
+
*whisper_params,
|
|
|
|
|
|
|
|
|
|
|
218 |
) -> Tuple[List[dict], float]:
|
219 |
"""
|
220 |
+
transcribe method for faster-whisper.
|
221 |
|
222 |
Parameters
|
223 |
----------
|
224 |
+
audio: Union[str, BinaryIO, np.ndarray]
|
225 |
Audio path or file binary or Audio numpy array
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
226 |
progress: gr.Progress
|
227 |
Indicator to show progress directly in gradio.
|
228 |
+
*whisper_params: tuple
|
229 |
+
Gradio components related to Whisper. see whisper_data_class.py for details.
|
230 |
|
231 |
Returns
|
232 |
----------
|
|
|
236 |
elapsed time for transcription
|
237 |
"""
|
238 |
start_time = time.time()
|
239 |
+
params = WhisperGradioComponents.to_values(*whisper_params)
|
240 |
+
|
241 |
+
if params.model_size != self.current_model_size or self.model is None or self.current_compute_type != params.compute_type:
|
242 |
+
self.update_model(params.model_size, params.compute_type, progress)
|
243 |
+
|
244 |
+
if params.lang == "Automatic Detection":
|
245 |
+
params.lang = None
|
246 |
|
247 |
def progress_callback(progress_value):
|
248 |
progress(progress_value, desc="Transcribing..")
|
249 |
|
|
|
|
|
|
|
|
|
250 |
segments_result = self.model.transcribe(audio=audio,
|
251 |
+
language=params.lang,
|
252 |
verbose=False,
|
253 |
+
beam_size=params.beam_size,
|
254 |
+
logprob_threshold=params.log_prob_threshold,
|
255 |
+
no_speech_threshold=params.no_speech_threshold,
|
256 |
+
task="translate" if params.is_translate and self.current_model_size in self.translatable_model else "transcribe",
|
257 |
+
fp16=True if params.compute_type == "float16" else False,
|
258 |
progress_callback=progress_callback)["segments"]
|
259 |
elapsed_time = time.time() - start_time
|
260 |
|
261 |
return segments_result, elapsed_time
|
262 |
|
263 |
+
def update_model(self,
|
264 |
+
model_size: str,
|
265 |
+
compute_type: str,
|
266 |
+
progress: gr.Progress,
|
267 |
+
):
|
268 |
"""
|
269 |
+
Update current model setting
|
270 |
+
|
271 |
+
Parameters
|
272 |
+
----------
|
273 |
+
model_size: str
|
274 |
+
Size of whisper model
|
275 |
+
compute_type: str
|
276 |
+
Compute type for transcription.
|
277 |
+
see more info : https://opennmt.net/CTranslate2/quantization.html
|
278 |
+
progress: gr.Progress
|
279 |
+
Indicator to show progress directly in gradio.
|
280 |
"""
|
281 |
+
progress(0, desc="Initializing Model..")
|
282 |
+
self.current_compute_type = compute_type
|
283 |
+
self.current_model_size = model_size
|
284 |
+
self.model = whisper.load_model(
|
285 |
+
name=model_size,
|
286 |
+
device=self.device,
|
287 |
+
download_root=os.path.join("models", "Whisper")
|
288 |
+
)
|
|
|
|
|
289 |
|
290 |
@staticmethod
|
291 |
def generate_and_write_file(file_name: str,
|
|
|
294 |
file_format: str,
|
295 |
) -> str:
|
296 |
"""
|
297 |
+
Writes subtitle file
|
298 |
+
|
299 |
+
Parameters
|
300 |
+
----------
|
301 |
+
file_name: str
|
302 |
+
Output file name
|
303 |
+
transcribed_segments: list
|
304 |
+
Text segments transcribed from audio
|
305 |
+
add_timestamp: bool
|
306 |
+
Determines whether to add a timestamp to the end of the filename.
|
307 |
+
file_format: str
|
308 |
+
File format to write. Supported formats: [SRT, WebVTT, txt]
|
309 |
+
|
310 |
+
Returns
|
311 |
+
----------
|
312 |
+
content: str
|
313 |
+
Result of the transcription
|
314 |
+
output_path: str
|
315 |
+
output file path
|
316 |
"""
|
317 |
timestamp = datetime.now().strftime("%m%d%H%M%S")
|
318 |
if add_timestamp:
|
|
|
338 |
|
339 |
@staticmethod
|
340 |
def format_time(elapsed_time: float) -> str:
|
341 |
+
"""
|
342 |
+
Get {hours} {minutes} {seconds} time format string
|
343 |
+
|
344 |
+
Parameters
|
345 |
+
----------
|
346 |
+
elapsed_time: str
|
347 |
+
Elapsed time for transcription
|
348 |
+
|
349 |
+
Returns
|
350 |
+
----------
|
351 |
+
Time format string
|
352 |
+
"""
|
353 |
hours, rem = divmod(elapsed_time, 3600)
|
354 |
minutes, seconds = divmod(rem, 60)
|
355 |
|