Spaces:
Running
Running
File size: 18,061 Bytes
63ab978 a63d493 63ab978 a63d493 63ab978 a63d493 63ab978 9cf2e86 91b9b83 2ff01cb 91b9b83 67cc6b1 63ab978 67cc6b1 5845bfc f9abd83 1f79d6e 00efe30 ae32f22 63ab978 9cf2e86 ccf78ae e29f6b4 00efe30 63ab978 9cf2e86 ccf78ae e29f6b4 00efe30 9cf2e86 91b9b83 6074f61 00efe30 6074f61 a63d493 6074f61 a63d493 e29f6b4 00efe30 6074f61 a63d493 6074f61 a63d493 6074f61 a63d493 6074f61 a63d493 6074f61 a63d493 6074f61 a63d493 6074f61 67cc6b1 63ab978 9cf2e86 ccf78ae e29f6b4 00efe30 6074f61 9cf2e86 ccf78ae e29f6b4 00efe30 9cf2e86 6074f61 00efe30 6074f61 21c25c6 a63d493 e29f6b4 00efe30 a63d493 91b9b83 21c25c6 6074f61 a63d493 6074f61 a63d493 6074f61 e29f6b4 91b9b83 9cf2e86 e29f6b4 00efe30 91b9b83 9cf2e86 e29f6b4 00efe30 9cf2e86 91b9b83 6074f61 00efe30 63ab978 a63d493 e29f6b4 00efe30 a63d493 6074f61 63ab978 a63d493 63ab978 a63d493 6074f61 a63d493 6074f61 67cc6b1 a63d493 e29f6b4 00efe30 a63d493 e29f6b4 00efe30 a63d493 e29f6b4 a63d493 00efe30 a63d493 00efe30 a63d493 00efe30 a63d493 00efe30 a63d493 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 |
import whisper
import gradio as gr
import time
import os
from typing import BinaryIO, Union, Tuple
import numpy as np
from datetime import datetime
import torch
from .base_interface import BaseInterface
from modules.subtitle_manager import get_srt, get_vtt, write_file, safe_filename
from modules.youtube_manager import get_ytdata, get_ytaudio
DEFAULT_MODEL_SIZE = "large-v2"
class WhisperInference(BaseInterface):
def __init__(self):
super().__init__()
self.current_model_size = None
self.model = None
self.available_models = whisper.available_models()
self.available_langs = sorted(list(whisper.tokenizer.LANGUAGES.values()))
self.device = "cuda" if torch.cuda.is_available() else "cpu"
self.available_compute_types = ["float16", "float32"]
self.current_compute_type = "float16" if self.device == "cuda" else "float32"
self.default_beam_size = 1
def transcribe_file(self,
fileobjs: list,
model_size: str,
lang: str,
subformat: str,
istranslate: bool,
add_timestamp: bool,
beam_size: int,
log_prob_threshold: float,
no_speech_threshold: float,
compute_type: str,
progress=gr.Progress()):
"""
Write subtitle file from Files
Parameters
----------
fileobjs: list
List of files to transcribe from gr.Files()
model_size: str
Whisper model size from gr.Dropdown()
lang: str
Source language of the file to transcribe from gr.Dropdown()
subformat: str
Subtitle format to write from gr.Dropdown(). Supported format: [SRT, WebVTT]
istranslate: bool
Boolean value from gr.Checkbox() that determines whether to translate to English.
It's Whisper's feature to translate speech from another language directly into English end-to-end.
add_timestamp: bool
Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the filename.
beam_size: int
Int value from gr.Number() that is used for decoding option.
log_prob_threshold: float
float value from gr.Number(). If the average log probability over sampled tokens is
below this value, treat as failed.
no_speech_threshold: float
float value from gr.Number(). If the no_speech probability is higher than this value AND
the average log probability over sampled tokens is below `log_prob_threshold`,
consider the segment as silent.
compute_type: str
compute type from gr.Dropdown().
progress: gr.Progress
Indicator to show progress directly in gradio.
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
"""
try:
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
files_info = {}
for fileobj in fileobjs:
progress(0, desc="Loading Audio..")
audio = whisper.load_audio(fileobj.name)
result, elapsed_time = self.transcribe(audio=audio,
lang=lang,
istranslate=istranslate,
beam_size=beam_size,
log_prob_threshold=log_prob_threshold,
no_speech_threshold=no_speech_threshold,
compute_type=compute_type,
progress=progress
)
progress(1, desc="Completed!")
file_name, file_ext = os.path.splitext(os.path.basename(fileobj.orig_name))
file_name = safe_filename(file_name)
subtitle = self.generate_and_write_subtitle(
file_name=file_name,
transcribed_segments=result,
add_timestamp=add_timestamp,
subformat=subformat
)
files_info[file_name] = {"subtitle": subtitle, "elapsed_time": elapsed_time}
total_result = ''
total_time = 0
for file_name, info in files_info.items():
total_result += '------------------------------------\n'
total_result += f'{file_name}\n\n'
total_result += f"{info['subtitle']}"
total_time += info["elapsed_time"]
return f"Done in {self.format_time(total_time)}! Subtitle is in the outputs folder.\n\n{total_result}"
except Exception as e:
print(f"Error transcribing file: {str(e)}")
return f"Error transcribing file: {str(e)}"
finally:
self.release_cuda_memory()
self.remove_input_files([fileobj.name for fileobj in fileobjs])
def transcribe_youtube(self,
youtubelink: str,
model_size: str,
lang: str,
subformat: str,
istranslate: bool,
add_timestamp: bool,
beam_size: int,
log_prob_threshold: float,
no_speech_threshold: float,
compute_type: str,
progress=gr.Progress()):
"""
Write subtitle file from Youtube
Parameters
----------
youtubelink: str
Link of Youtube to transcribe from gr.Textbox()
model_size: str
Whisper model size from gr.Dropdown()
lang: str
Source language of the file to transcribe from gr.Dropdown()
subformat: str
Subtitle format to write from gr.Dropdown(). Supported format: [SRT, WebVTT]
istranslate: bool
Boolean value from gr.Checkbox() that determines whether to translate to English.
It's Whisper's feature to translate speech from another language directly into English end-to-end.
add_timestamp: bool
Boolean value from gr.Checkbox() that determines whether to add a timestamp at the end of the filename.
beam_size: int
Int value from gr.Number() that is used for decoding option.
log_prob_threshold: float
float value from gr.Number(). If the average log probability over sampled tokens is
below this value, treat as failed.
no_speech_threshold: float
float value from gr.Number(). If the no_speech probability is higher than this value AND
the average log probability over sampled tokens is below `log_prob_threshold`,
consider the segment as silent.
compute_type: str
compute type from gr.Dropdown().
progress: gr.Progress
Indicator to show progress directly in gradio.
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
"""
try:
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
progress(0, desc="Loading Audio from Youtube..")
yt = get_ytdata(youtubelink)
audio = whisper.load_audio(get_ytaudio(yt))
result, elapsed_time = self.transcribe(audio=audio,
lang=lang,
istranslate=istranslate,
beam_size=beam_size,
log_prob_threshold=log_prob_threshold,
no_speech_threshold=no_speech_threshold,
compute_type=compute_type,
progress=progress)
progress(1, desc="Completed!")
file_name = safe_filename(yt.title)
subtitle = self.generate_and_write_subtitle(
file_name=file_name,
transcribed_segments=result,
add_timestamp=add_timestamp,
subformat=subformat
)
return f"Done in {self.format_time(elapsed_time)}! Subtitle file is in the outputs folder.\n\n{subtitle}"
except Exception as e:
print(f"Error transcribing youtube video: {str(e)}")
return f"Error transcribing youtube video: {str(e)}"
finally:
try:
if 'yt' not in locals():
yt = get_ytdata(youtubelink)
file_path = get_ytaudio(yt)
else:
file_path = get_ytaudio(yt)
self.release_cuda_memory()
self.remove_input_files([file_path])
except Exception as cleanup_error:
pass
def transcribe_mic(self,
micaudio: str,
model_size: str,
lang: str,
subformat: str,
istranslate: bool,
beam_size: int,
log_prob_threshold: float,
no_speech_threshold: float,
compute_type: str,
progress=gr.Progress()):
"""
Write subtitle file from microphone
Parameters
----------
micaudio: str
Audio file path from gr.Microphone()
model_size: str
Whisper model size from gr.Dropdown()
lang: str
Source language of the file to transcribe from gr.Dropdown()
subformat: str
Subtitle format to write from gr.Dropdown(). Supported format: [SRT, WebVTT]
istranslate: bool
Boolean value from gr.Checkbox() that determines whether to translate to English.
It's Whisper's feature to translate speech from another language directly into English end-to-end.
beam_size: int
Int value from gr.Number() that is used for decoding option.
log_prob_threshold: float
float value from gr.Number(). If the average log probability over sampled tokens is
below this value, treat as failed.
no_speech_threshold: float
float value from gr.Number(). If the no_speech probability is higher than this value AND
the average log probability over sampled tokens is below `log_prob_threshold`,
consider the segment as silent.
compute_type: str
compute type from gr.Dropdown().
progress: gr.Progress
Indicator to show progress directly in gradio.
I use a forked version of whisper for this. To see more info : https://github.com/jhj0517/jhj0517-whisper/tree/add-progress-callback
"""
try:
self.update_model_if_needed(model_size=model_size, compute_type=compute_type, progress=progress)
result, elapsed_time = self.transcribe(audio=micaudio,
lang=lang,
istranslate=istranslate,
beam_size=beam_size,
log_prob_threshold=log_prob_threshold,
no_speech_threshold=no_speech_threshold,
compute_type=compute_type,
progress=progress)
progress(1, desc="Completed!")
subtitle = self.generate_and_write_subtitle(
file_name="Mic",
transcribed_segments=result,
add_timestamp=True,
subformat=subformat
)
return f"Done in {self.format_time(elapsed_time)}! Subtitle file is in the outputs folder.\n\n{subtitle}"
except Exception as e:
print(f"Error transcribing mic: {str(e)}")
return f"Error transcribing mic: {str(e)}"
finally:
self.release_cuda_memory()
self.remove_input_files([micaudio])
def transcribe(self,
audio: Union[str, np.ndarray, torch.Tensor],
lang: str,
istranslate: bool,
beam_size: int,
log_prob_threshold: float,
no_speech_threshold: float,
compute_type: str,
progress: gr.Progress
) -> Tuple[list[dict], float]:
"""
transcribe method for OpenAI's Whisper implementation.
Parameters
----------
audio: Union[str, BinaryIO, torch.Tensor]
Audio path or file binary or Audio numpy array
lang: str
Source language of the file to transcribe from gr.Dropdown()
istranslate: bool
Boolean value from gr.Checkbox() that determines whether to translate to English.
It's Whisper's feature to translate speech from another language directly into English end-to-end.
beam_size: int
Int value from gr.Number() that is used for decoding option.
log_prob_threshold: float
float value from gr.Number(). If the average log probability over sampled tokens is
below this value, treat as failed.
no_speech_threshold: float
float value from gr.Number(). If the no_speech probability is higher than this value AND
the average log probability over sampled tokens is below `log_prob_threshold`,
consider the segment as silent.
compute_type: str
compute type from gr.Dropdown().
progress: gr.Progress
Indicator to show progress directly in gradio.
Returns
----------
segments_result: list[dict]
list of dicts that includes start, end timestamps and transcribed text
elapsed_time: float
elapsed time for transcription
"""
start_time = time.time()
def progress_callback(progress_value):
progress(progress_value, desc="Transcribing..")
if lang == "Automatic Detection":
lang = None
translatable_model = ["large", "large-v1", "large-v2"]
segments_result = self.model.transcribe(audio=audio,
language=lang,
verbose=False,
beam_size=beam_size,
logprob_threshold=log_prob_threshold,
no_speech_threshold=no_speech_threshold,
task="translate" if istranslate and self.current_model_size in translatable_model else "transcribe",
fp16=True if compute_type == "float16" else False,
progress_callback=progress_callback)["segments"]
elapsed_time = time.time() - start_time
return segments_result, elapsed_time
def update_model_if_needed(self,
model_size: str,
compute_type: str,
progress: gr.Progress,
):
"""
Initialize model if it doesn't match with current model setting
"""
if compute_type != self.current_compute_type:
self.current_compute_type = compute_type
if model_size != self.current_model_size or self.model is None:
progress(0, desc="Initializing Model..")
self.current_model_size = model_size
self.model = whisper.load_model(
name=model_size,
device=self.device,
download_root=os.path.join("models", "Whisper")
)
@staticmethod
def generate_and_write_subtitle(file_name: str,
transcribed_segments: list,
add_timestamp: bool,
subformat: str,
) -> str:
"""
This method writes subtitle file and returns str to gr.Textbox
"""
timestamp = datetime.now().strftime("%m%d%H%M%S")
if add_timestamp:
output_path = os.path.join("outputs", f"{file_name}-{timestamp}")
else:
output_path = os.path.join("outputs", f"{file_name}")
if subformat == "SRT":
subtitle = get_srt(transcribed_segments)
write_file(subtitle, f"{output_path}.srt")
elif subformat == "WebVTT":
subtitle = get_vtt(transcribed_segments)
write_file(subtitle, f"{output_path}.vtt")
return subtitle
@staticmethod
def format_time(elapsed_time: float) -> str:
hours, rem = divmod(elapsed_time, 3600)
minutes, seconds = divmod(rem, 60)
time_str = ""
if hours:
time_str += f"{hours} hours "
if minutes:
time_str += f"{minutes} minutes "
seconds = round(seconds)
time_str += f"{seconds} seconds"
return time_str.strip()
|