Spaces:
Running
Running
jhj0517
commited on
Commit
Β·
00efe30
1
Parent(s):
f962fd9
add compute_type dropdown
Browse files- app.py +6 -3
- modules/faster_whisper_inference.py +34 -21
- modules/whisper_Inference.py +39 -15
app.py
CHANGED
@@ -59,6 +59,7 @@ class App:
|
|
59 |
nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
|
60 |
nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
|
61 |
nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
|
|
|
62 |
with gr.Row():
|
63 |
btn_run = gr.Button("GENERATE SUBTITLE FILE", variant="primary")
|
64 |
with gr.Row():
|
@@ -66,7 +67,7 @@ class App:
|
|
66 |
btn_openfolder = gr.Button('π', scale=2)
|
67 |
|
68 |
params = [input_file, dd_model, dd_lang, dd_subformat, cb_translate, cb_timestamp]
|
69 |
-
advanced_params = [nb_beam_size, nb_log_prob_threshold, nb_no_speech_threshold]
|
70 |
btn_run.click(fn=self.whisper_inf.transcribe_file,
|
71 |
inputs=params + advanced_params,
|
72 |
outputs=[tb_indicator])
|
@@ -97,6 +98,7 @@ class App:
|
|
97 |
nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
|
98 |
nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
|
99 |
nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
|
|
|
100 |
with gr.Row():
|
101 |
btn_run = gr.Button("GENERATE SUBTITLE FILE", variant="primary")
|
102 |
with gr.Row():
|
@@ -104,7 +106,7 @@ class App:
|
|
104 |
btn_openfolder = gr.Button('π', scale=2)
|
105 |
|
106 |
params = [tb_youtubelink, dd_model, dd_lang, dd_subformat, cb_translate, cb_timestamp]
|
107 |
-
advanced_params = [nb_beam_size, nb_log_prob_threshold, nb_no_speech_threshold]
|
108 |
btn_run.click(fn=self.whisper_inf.transcribe_youtube,
|
109 |
inputs=params + advanced_params,
|
110 |
outputs=[tb_indicator])
|
@@ -128,6 +130,7 @@ class App:
|
|
128 |
nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
|
129 |
nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
|
130 |
nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
|
|
|
131 |
with gr.Row():
|
132 |
btn_run = gr.Button("GENERATE SUBTITLE FILE", variant="primary")
|
133 |
with gr.Row():
|
@@ -135,7 +138,7 @@ class App:
|
|
135 |
btn_openfolder = gr.Button('π', scale=2)
|
136 |
|
137 |
params = [mic_input, dd_model, dd_lang, dd_subformat, cb_translate]
|
138 |
-
advanced_params = [nb_beam_size, nb_log_prob_threshold, nb_no_speech_threshold]
|
139 |
btn_run.click(fn=self.whisper_inf.transcribe_mic,
|
140 |
inputs=params + advanced_params,
|
141 |
outputs=[tb_indicator])
|
|
|
59 |
nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
|
60 |
nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
|
61 |
nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
|
62 |
+
dd_compute_type = gr.Dropdown(label="Compute Type", choices=self.whisper_inf.available_compute_types, value=self.whisper_inf.current_compute_type, interactive=True)
|
63 |
with gr.Row():
|
64 |
btn_run = gr.Button("GENERATE SUBTITLE FILE", variant="primary")
|
65 |
with gr.Row():
|
|
|
67 |
btn_openfolder = gr.Button('π', scale=2)
|
68 |
|
69 |
params = [input_file, dd_model, dd_lang, dd_subformat, cb_translate, cb_timestamp]
|
70 |
+
advanced_params = [nb_beam_size, nb_log_prob_threshold, nb_no_speech_threshold, dd_compute_type]
|
71 |
btn_run.click(fn=self.whisper_inf.transcribe_file,
|
72 |
inputs=params + advanced_params,
|
73 |
outputs=[tb_indicator])
|
|
|
98 |
nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
|
99 |
nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
|
100 |
nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
|
101 |
+
dd_compute_type = gr.Dropdown(label="Compute Type", choices=self.whisper_inf.available_compute_types, value=self.whisper_inf.current_compute_type, interactive=True)
|
102 |
with gr.Row():
|
103 |
btn_run = gr.Button("GENERATE SUBTITLE FILE", variant="primary")
|
104 |
with gr.Row():
|
|
|
106 |
btn_openfolder = gr.Button('π', scale=2)
|
107 |
|
108 |
params = [tb_youtubelink, dd_model, dd_lang, dd_subformat, cb_translate, cb_timestamp]
|
109 |
+
advanced_params = [nb_beam_size, nb_log_prob_threshold, nb_no_speech_threshold, dd_compute_type]
|
110 |
btn_run.click(fn=self.whisper_inf.transcribe_youtube,
|
111 |
inputs=params + advanced_params,
|
112 |
outputs=[tb_indicator])
|
|
|
130 |
nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
|
131 |
nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
|
132 |
nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
|
133 |
+
dd_compute_type = gr.Dropdown(label="Compute Type", choices=self.whisper_inf.available_compute_types, value=self.whisper_inf.current_compute_type, interactive=True)
|
134 |
with gr.Row():
|
135 |
btn_run = gr.Button("GENERATE SUBTITLE FILE", variant="primary")
|
136 |
with gr.Row():
|
|
|
138 |
btn_openfolder = gr.Button('π', scale=2)
|
139 |
|
140 |
params = [mic_input, dd_model, dd_lang, dd_subformat, cb_translate]
|
141 |
+
advanced_params = [nb_beam_size, nb_log_prob_threshold, nb_no_speech_threshold, dd_compute_type]
|
142 |
btn_run.click(fn=self.whisper_inf.transcribe_mic,
|
143 |
inputs=params + advanced_params,
|
144 |
outputs=[tb_indicator])
|
modules/faster_whisper_inference.py
CHANGED
@@ -24,9 +24,10 @@ class FasterWhisperInference(BaseInterface):
|
|
24 |
self.available_models = whisper.available_models()
|
25 |
self.available_langs = sorted(list(whisper.tokenizer.LANGUAGES.values()))
|
26 |
self.translatable_models = ["large", "large-v1", "large-v2"]
|
27 |
-
self.default_beam_size = 1
|
28 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
29 |
-
self.
|
|
|
|
|
30 |
|
31 |
def transcribe_file(self,
|
32 |
fileobjs: list,
|
@@ -38,6 +39,7 @@ class FasterWhisperInference(BaseInterface):
|
|
38 |
beam_size: int,
|
39 |
log_prob_threshold: float,
|
40 |
no_speech_threshold: float,
|
|
|
41 |
progress=gr.Progress()
|
42 |
) -> str:
|
43 |
"""
|
@@ -67,6 +69,9 @@ class FasterWhisperInference(BaseInterface):
|
|
67 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
68 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
69 |
consider the segment as silent.
|
|
|
|
|
|
|
70 |
progress: gr.Progress
|
71 |
Indicator to show progress directly in gradio.
|
72 |
|
@@ -75,8 +80,7 @@ class FasterWhisperInference(BaseInterface):
|
|
75 |
String to return to gr.Textbox()
|
76 |
"""
|
77 |
try:
|
78 |
-
|
79 |
-
self.initialize_model(model_size=model_size, progress=progress)
|
80 |
|
81 |
if lang == "Automatic Detection":
|
82 |
lang = None
|
@@ -129,6 +133,7 @@ class FasterWhisperInference(BaseInterface):
|
|
129 |
beam_size: int,
|
130 |
log_prob_threshold: float,
|
131 |
no_speech_threshold: float,
|
|
|
132 |
progress=gr.Progress()
|
133 |
) -> str:
|
134 |
"""
|
@@ -158,6 +163,9 @@ class FasterWhisperInference(BaseInterface):
|
|
158 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
159 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
160 |
consider the segment as silent.
|
|
|
|
|
|
|
161 |
progress: gr.Progress
|
162 |
Indicator to show progress directly in gradio.
|
163 |
|
@@ -166,8 +174,7 @@ class FasterWhisperInference(BaseInterface):
|
|
166 |
String to return to gr.Textbox()
|
167 |
"""
|
168 |
try:
|
169 |
-
|
170 |
-
self.initialize_model(model_size=model_size, progress=progress)
|
171 |
|
172 |
if lang == "Automatic Detection":
|
173 |
lang = None
|
@@ -220,6 +227,7 @@ class FasterWhisperInference(BaseInterface):
|
|
220 |
beam_size: int,
|
221 |
log_prob_threshold: float,
|
222 |
no_speech_threshold: float,
|
|
|
223 |
progress=gr.Progress()
|
224 |
) -> str:
|
225 |
"""
|
@@ -246,6 +254,9 @@ class FasterWhisperInference(BaseInterface):
|
|
246 |
no_speech_threshold: float
|
247 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
248 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
|
|
|
|
|
|
249 |
consider the segment as silent.
|
250 |
progress: gr.Progress
|
251 |
Indicator to show progress directly in gradio.
|
@@ -255,8 +266,7 @@ class FasterWhisperInference(BaseInterface):
|
|
255 |
String to return to gr.Textbox()
|
256 |
"""
|
257 |
try:
|
258 |
-
|
259 |
-
self.initialize_model(model_size=model_size, progress=progress)
|
260 |
|
261 |
if lang == "Automatic Detection":
|
262 |
lang = None
|
@@ -353,21 +363,24 @@ class FasterWhisperInference(BaseInterface):
|
|
353 |
elapsed_time = time.time() - start_time
|
354 |
return segments_result, elapsed_time
|
355 |
|
356 |
-
def
|
357 |
-
|
358 |
-
|
359 |
-
|
|
|
360 |
"""
|
361 |
-
Initialize model if it doesn't match with current model
|
362 |
"""
|
363 |
-
|
364 |
-
|
365 |
-
|
366 |
-
|
367 |
-
|
368 |
-
|
369 |
-
|
370 |
-
|
|
|
|
|
371 |
|
372 |
@staticmethod
|
373 |
def generate_and_write_subtitle(file_name: str,
|
|
|
24 |
self.available_models = whisper.available_models()
|
25 |
self.available_langs = sorted(list(whisper.tokenizer.LANGUAGES.values()))
|
26 |
self.translatable_models = ["large", "large-v1", "large-v2"]
|
|
|
27 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
28 |
+
self.available_compute_types = ["int8", "int8_float32", "int8_float16", "int8_bfloat16", "int16", "float16", "bfloat16", "float32"]
|
29 |
+
self.current_compute_type = "float16" if self.device == "cuda" else "float32"
|
30 |
+
self.default_beam_size = 1
|
31 |
|
32 |
def transcribe_file(self,
|
33 |
fileobjs: list,
|
|
|
39 |
beam_size: int,
|
40 |
log_prob_threshold: float,
|
41 |
no_speech_threshold: float,
|
42 |
+
compute_type: str,
|
43 |
progress=gr.Progress()
|
44 |
) -> str:
|
45 |
"""
|
|
|
69 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
70 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
71 |
consider the segment as silent.
|
72 |
+
compute_type: str
|
73 |
+
compute type from gr.Dropdown().
|
74 |
+
see more info : https://opennmt.net/CTranslate2/quantization.html
|
75 |
progress: gr.Progress
|
76 |
Indicator to show progress directly in gradio.
|
77 |
|
|
|
80 |
String to return to gr.Textbox()
|
81 |
"""
|
82 |
try:
|
83 |
+
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
|
|
|
84 |
|
85 |
if lang == "Automatic Detection":
|
86 |
lang = None
|
|
|
133 |
beam_size: int,
|
134 |
log_prob_threshold: float,
|
135 |
no_speech_threshold: float,
|
136 |
+
compute_type: str,
|
137 |
progress=gr.Progress()
|
138 |
) -> str:
|
139 |
"""
|
|
|
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 |
+
see more info : https://opennmt.net/CTranslate2/quantization.html
|
169 |
progress: gr.Progress
|
170 |
Indicator to show progress directly in gradio.
|
171 |
|
|
|
174 |
String to return to gr.Textbox()
|
175 |
"""
|
176 |
try:
|
177 |
+
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
|
|
|
178 |
|
179 |
if lang == "Automatic Detection":
|
180 |
lang = None
|
|
|
227 |
beam_size: int,
|
228 |
log_prob_threshold: float,
|
229 |
no_speech_threshold: float,
|
230 |
+
compute_type: str,
|
231 |
progress=gr.Progress()
|
232 |
) -> str:
|
233 |
"""
|
|
|
254 |
no_speech_threshold: float
|
255 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
256 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
257 |
+
compute_type: str
|
258 |
+
compute type from gr.Dropdown().
|
259 |
+
see more info : https://opennmt.net/CTranslate2/quantization.html
|
260 |
consider the segment as silent.
|
261 |
progress: gr.Progress
|
262 |
Indicator to show progress directly in gradio.
|
|
|
266 |
String to return to gr.Textbox()
|
267 |
"""
|
268 |
try:
|
269 |
+
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
|
|
|
270 |
|
271 |
if lang == "Automatic Detection":
|
272 |
lang = None
|
|
|
363 |
elapsed_time = time.time() - start_time
|
364 |
return segments_result, elapsed_time
|
365 |
|
366 |
+
def update_model_if_needed(self,
|
367 |
+
model_size: str,
|
368 |
+
compute_type: str,
|
369 |
+
progress: gr.Progress
|
370 |
+
):
|
371 |
"""
|
372 |
+
Initialize model if it doesn't match with current model setting
|
373 |
"""
|
374 |
+
if model_size != self.current_model_size or self.model is None or self.current_compute_type != compute_type:
|
375 |
+
progress(0, desc="Initializing Model..")
|
376 |
+
self.current_model_size = model_size
|
377 |
+
self.current_compute_type = compute_type
|
378 |
+
self.model = faster_whisper.WhisperModel(
|
379 |
+
device=self.device,
|
380 |
+
model_size_or_path=model_size,
|
381 |
+
download_root=os.path.join("models", "Whisper", "faster-whisper"),
|
382 |
+
compute_type=self.current_compute_type
|
383 |
+
)
|
384 |
|
385 |
@staticmethod
|
386 |
def generate_and_write_subtitle(file_name: str,
|
modules/whisper_Inference.py
CHANGED
@@ -22,6 +22,8 @@ class WhisperInference(BaseInterface):
|
|
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.default_beam_size = 1
|
26 |
|
27 |
def transcribe_file(self,
|
@@ -34,6 +36,7 @@ class WhisperInference(BaseInterface):
|
|
34 |
beam_size: int,
|
35 |
log_prob_threshold: float,
|
36 |
no_speech_threshold: float,
|
|
|
37 |
progress=gr.Progress()):
|
38 |
"""
|
39 |
Write subtitle file from Files
|
@@ -62,14 +65,15 @@ class WhisperInference(BaseInterface):
|
|
62 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
63 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
64 |
consider the segment as silent.
|
|
|
|
|
65 |
progress: gr.Progress
|
66 |
Indicator to show progress directly in gradio.
|
67 |
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
|
68 |
"""
|
69 |
|
70 |
try:
|
71 |
-
|
72 |
-
self.initialize_model(model_size=model_size, progress=progress)
|
73 |
|
74 |
files_info = {}
|
75 |
for fileobj in fileobjs:
|
@@ -82,7 +86,9 @@ class WhisperInference(BaseInterface):
|
|
82 |
beam_size=beam_size,
|
83 |
log_prob_threshold=log_prob_threshold,
|
84 |
no_speech_threshold=no_speech_threshold,
|
85 |
-
|
|
|
|
|
86 |
progress(1, desc="Completed!")
|
87 |
|
88 |
file_name, file_ext = os.path.splitext(os.path.basename(fileobj.orig_name))
|
@@ -122,6 +128,7 @@ class WhisperInference(BaseInterface):
|
|
122 |
beam_size: int,
|
123 |
log_prob_threshold: float,
|
124 |
no_speech_threshold: float,
|
|
|
125 |
progress=gr.Progress()):
|
126 |
"""
|
127 |
Write subtitle file from Youtube
|
@@ -150,13 +157,14 @@ class WhisperInference(BaseInterface):
|
|
150 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
151 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
152 |
consider the segment as silent.
|
|
|
|
|
153 |
progress: gr.Progress
|
154 |
Indicator to show progress directly in gradio.
|
155 |
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
|
156 |
"""
|
157 |
try:
|
158 |
-
|
159 |
-
self.initialize_model(model_size=model_size, progress=progress)
|
160 |
|
161 |
progress(0, desc="Loading Audio from Youtube..")
|
162 |
yt = get_ytdata(youtubelink)
|
@@ -168,6 +176,7 @@ class WhisperInference(BaseInterface):
|
|
168 |
beam_size=beam_size,
|
169 |
log_prob_threshold=log_prob_threshold,
|
170 |
no_speech_threshold=no_speech_threshold,
|
|
|
171 |
progress=progress)
|
172 |
progress(1, desc="Completed!")
|
173 |
|
@@ -205,6 +214,7 @@ class WhisperInference(BaseInterface):
|
|
205 |
beam_size: int,
|
206 |
log_prob_threshold: float,
|
207 |
no_speech_threshold: float,
|
|
|
208 |
progress=gr.Progress()):
|
209 |
"""
|
210 |
Write subtitle file from microphone
|
@@ -231,14 +241,15 @@ class WhisperInference(BaseInterface):
|
|
231 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
232 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
233 |
consider the segment as silent.
|
|
|
|
|
234 |
progress: gr.Progress
|
235 |
Indicator to show progress directly in gradio.
|
236 |
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
|
237 |
"""
|
238 |
|
239 |
try:
|
240 |
-
|
241 |
-
self.initialize_model(model_size=model_size, progress=progress)
|
242 |
|
243 |
result, elapsed_time = self.transcribe(audio=micaudio,
|
244 |
lang=lang,
|
@@ -246,6 +257,7 @@ class WhisperInference(BaseInterface):
|
|
246 |
beam_size=beam_size,
|
247 |
log_prob_threshold=log_prob_threshold,
|
248 |
no_speech_threshold=no_speech_threshold,
|
|
|
249 |
progress=progress)
|
250 |
progress(1, desc="Completed!")
|
251 |
|
@@ -271,6 +283,7 @@ class WhisperInference(BaseInterface):
|
|
271 |
beam_size: int,
|
272 |
log_prob_threshold: float,
|
273 |
no_speech_threshold: float,
|
|
|
274 |
progress: gr.Progress
|
275 |
) -> Tuple[list[dict], float]:
|
276 |
"""
|
@@ -294,6 +307,8 @@ class WhisperInference(BaseInterface):
|
|
294 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
295 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
296 |
consider the segment as silent.
|
|
|
|
|
297 |
progress: gr.Progress
|
298 |
Indicator to show progress directly in gradio.
|
299 |
|
@@ -320,21 +335,30 @@ class WhisperInference(BaseInterface):
|
|
320 |
logprob_threshold=log_prob_threshold,
|
321 |
no_speech_threshold=no_speech_threshold,
|
322 |
task="translate" if istranslate and self.current_model_size in translatable_model else "transcribe",
|
|
|
323 |
progress_callback=progress_callback)["segments"]
|
324 |
elapsed_time = time.time() - start_time
|
325 |
|
326 |
return segments_result, elapsed_time
|
327 |
|
328 |
-
def
|
329 |
-
|
330 |
-
|
331 |
-
|
|
|
332 |
"""
|
333 |
-
Initialize model if it doesn't match with current model
|
334 |
"""
|
335 |
-
|
336 |
-
|
337 |
-
self.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
338 |
|
339 |
@staticmethod
|
340 |
def generate_and_write_subtitle(file_name: str,
|
|
|
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,
|
|
|
36 |
beam_size: int,
|
37 |
log_prob_threshold: float,
|
38 |
no_speech_threshold: float,
|
39 |
+
compute_type: str,
|
40 |
progress=gr.Progress()):
|
41 |
"""
|
42 |
Write subtitle file from Files
|
|
|
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 |
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
|
73 |
"""
|
74 |
|
75 |
try:
|
76 |
+
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
|
|
|
77 |
|
78 |
files_info = {}
|
79 |
for fileobj in fileobjs:
|
|
|
86 |
beam_size=beam_size,
|
87 |
log_prob_threshold=log_prob_threshold,
|
88 |
no_speech_threshold=no_speech_threshold,
|
89 |
+
compute_type=compute_type,
|
90 |
+
progress=progress
|
91 |
+
)
|
92 |
progress(1, desc="Completed!")
|
93 |
|
94 |
file_name, file_ext = os.path.splitext(os.path.basename(fileobj.orig_name))
|
|
|
128 |
beam_size: int,
|
129 |
log_prob_threshold: float,
|
130 |
no_speech_threshold: float,
|
131 |
+
compute_type: str,
|
132 |
progress=gr.Progress()):
|
133 |
"""
|
134 |
Write subtitle file from Youtube
|
|
|
157 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
158 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
159 |
consider the segment as silent.
|
160 |
+
compute_type: str
|
161 |
+
compute type from gr.Dropdown().
|
162 |
progress: gr.Progress
|
163 |
Indicator to show progress directly in gradio.
|
164 |
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
|
165 |
"""
|
166 |
try:
|
167 |
+
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
|
|
|
168 |
|
169 |
progress(0, desc="Loading Audio from Youtube..")
|
170 |
yt = get_ytdata(youtubelink)
|
|
|
176 |
beam_size=beam_size,
|
177 |
log_prob_threshold=log_prob_threshold,
|
178 |
no_speech_threshold=no_speech_threshold,
|
179 |
+
compute_type=compute_type,
|
180 |
progress=progress)
|
181 |
progress(1, desc="Completed!")
|
182 |
|
|
|
214 |
beam_size: int,
|
215 |
log_prob_threshold: float,
|
216 |
no_speech_threshold: float,
|
217 |
+
compute_type: str,
|
218 |
progress=gr.Progress()):
|
219 |
"""
|
220 |
Write subtitle file from microphone
|
|
|
241 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
242 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
243 |
consider the segment as silent.
|
244 |
+
compute_type: str
|
245 |
+
compute type from gr.Dropdown().
|
246 |
progress: gr.Progress
|
247 |
Indicator to show progress directly in gradio.
|
248 |
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
|
249 |
"""
|
250 |
|
251 |
try:
|
252 |
+
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
|
|
|
253 |
|
254 |
result, elapsed_time = self.transcribe(audio=micaudio,
|
255 |
lang=lang,
|
|
|
257 |
beam_size=beam_size,
|
258 |
log_prob_threshold=log_prob_threshold,
|
259 |
no_speech_threshold=no_speech_threshold,
|
260 |
+
compute_type=compute_type,
|
261 |
progress=progress)
|
262 |
progress(1, desc="Completed!")
|
263 |
|
|
|
283 |
beam_size: int,
|
284 |
log_prob_threshold: float,
|
285 |
no_speech_threshold: float,
|
286 |
+
compute_type: str,
|
287 |
progress: gr.Progress
|
288 |
) -> Tuple[list[dict], float]:
|
289 |
"""
|
|
|
307 |
float value from gr.Number(). If the no_speech probability is higher than this value AND
|
308 |
the average log probability over sampled tokens is below `log_prob_threshold`,
|
309 |
consider the segment as silent.
|
310 |
+
compute_type: str
|
311 |
+
compute type from gr.Dropdown().
|
312 |
progress: gr.Progress
|
313 |
Indicator to show progress directly in gradio.
|
314 |
|
|
|
335 |
logprob_threshold=log_prob_threshold,
|
336 |
no_speech_threshold=no_speech_threshold,
|
337 |
task="translate" if istranslate and self.current_model_size in translatable_model else "transcribe",
|
338 |
+
fp16=True if compute_type == "float16" else False,
|
339 |
progress_callback=progress_callback)["segments"]
|
340 |
elapsed_time = time.time() - start_time
|
341 |
|
342 |
return segments_result, elapsed_time
|
343 |
|
344 |
+
def update_model_if_needed(self,
|
345 |
+
model_size: str,
|
346 |
+
compute_type: str,
|
347 |
+
progress: gr.Progress,
|
348 |
+
):
|
349 |
"""
|
350 |
+
Initialize model if it doesn't match with current model setting
|
351 |
"""
|
352 |
+
if compute_type != self.current_compute_type:
|
353 |
+
self.current_compute_type = compute_type
|
354 |
+
if model_size != self.current_model_size or self.model is None:
|
355 |
+
progress(0, desc="Initializing Model..")
|
356 |
+
self.current_model_size = model_size
|
357 |
+
self.model = whisper.load_model(
|
358 |
+
name=model_size,
|
359 |
+
device=self.device,
|
360 |
+
download_root=os.path.join("models", "Whisper")
|
361 |
+
)
|
362 |
|
363 |
@staticmethod
|
364 |
def generate_and_write_subtitle(file_name: str,
|