Spaces:
Runtime error
Runtime error
Injected gradio-lite via gr.HTML(), added WebGPU support
Browse files
app.py
CHANGED
@@ -1,781 +1,109 @@
|
|
1 |
-
|
2 |
-
import json
|
3 |
-
import math
|
4 |
-
from typing import Callable, Iterator, Union
|
5 |
-
import argparse
|
6 |
-
|
7 |
-
from io import StringIO
|
8 |
import os
|
9 |
-
import pathlib
|
10 |
import tempfile
|
11 |
-
import zipfile
|
12 |
-
import numpy as np
|
13 |
-
|
14 |
-
import torch
|
15 |
-
|
16 |
-
from src.config import VAD_INITIAL_PROMPT_MODE_VALUES, ApplicationConfig, VadInitialPromptMode
|
17 |
-
from src.diarization.diarization import Diarization
|
18 |
-
from src.diarization.diarizationContainer import DiarizationContainer
|
19 |
-
from src.diarization.transcriptLoader import load_transcript
|
20 |
-
from src.hooks.progressListener import ProgressListener
|
21 |
-
from src.hooks.subTaskProgressListener import SubTaskProgressListener
|
22 |
-
from src.languages import get_language_names
|
23 |
-
from src.modelCache import ModelCache
|
24 |
-
from src.prompts.jsonPromptStrategy import JsonPromptStrategy
|
25 |
-
from src.prompts.prependPromptStrategy import PrependPromptStrategy
|
26 |
-
from src.source import AudioSource, get_audio_source_collection
|
27 |
-
from src.vadParallel import ParallelContext, ParallelTranscription
|
28 |
-
|
29 |
-
# External programs
|
30 |
import ffmpeg
|
31 |
-
|
32 |
-
|
33 |
-
import
|
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 |
-
self.vad_cpu_cores = min(os.cpu_count(), MAX_AUTO_CPU_CORES)
|
93 |
-
print("[Auto parallel] Using GPU devices " + str(self.parallel_device_list) + " and " + str(self.vad_cpu_cores) + " CPU cores for VAD/transcription.")
|
94 |
-
|
95 |
-
def set_diarization(self, auth_token: str, enable_daemon_process: bool = True, **kwargs):
|
96 |
-
if self.diarization is None:
|
97 |
-
self.diarization = DiarizationContainer(auth_token=auth_token, enable_daemon_process=enable_daemon_process,
|
98 |
-
auto_cleanup_timeout_seconds=self.app_config.diarization_process_timeout,
|
99 |
-
cache=self.model_cache)
|
100 |
-
# Set parameters
|
101 |
-
self.diarization_kwargs = kwargs
|
102 |
-
|
103 |
-
def unset_diarization(self):
|
104 |
-
if self.diarization is not None:
|
105 |
-
self.diarization.cleanup()
|
106 |
-
self.diarization_kwargs = None
|
107 |
-
|
108 |
-
# Entry function for the simple tab
|
109 |
-
def transcribe_webui_simple(self, modelName, languageName, urlData, multipleFiles, microphoneData, task,
|
110 |
-
vad, vadMergeWindow, vadMaxMergeSize,
|
111 |
-
word_timestamps: bool = False, highlight_words: bool = False,
|
112 |
-
diarization: bool = False, diarization_speakers: int = 2):
|
113 |
-
return self.transcribe_webui_simple_progress(modelName, languageName, urlData, multipleFiles, microphoneData, task,
|
114 |
-
vad, vadMergeWindow, vadMaxMergeSize,
|
115 |
-
word_timestamps, highlight_words,
|
116 |
-
diarization, diarization_speakers)
|
117 |
-
|
118 |
-
# Entry function for the simple tab progress
|
119 |
-
def transcribe_webui_simple_progress(self, modelName, languageName, urlData, multipleFiles, microphoneData, task,
|
120 |
-
vad, vadMergeWindow, vadMaxMergeSize,
|
121 |
-
word_timestamps: bool = False, highlight_words: bool = False,
|
122 |
-
diarization: bool = False, diarization_speakers: int = 2,
|
123 |
-
progress=gr.Progress()):
|
124 |
-
|
125 |
-
vadOptions = VadOptions(vad, vadMergeWindow, vadMaxMergeSize, self.app_config.vad_padding, self.app_config.vad_prompt_window, self.app_config.vad_initial_prompt_mode)
|
126 |
-
|
127 |
-
if diarization:
|
128 |
-
self.set_diarization(auth_token=self.app_config.auth_token, num_speakers=diarization_speakers)
|
129 |
-
else:
|
130 |
-
self.unset_diarization()
|
131 |
-
|
132 |
-
return self.transcribe_webui(modelName, languageName, urlData, multipleFiles, microphoneData, task, vadOptions,
|
133 |
-
word_timestamps=word_timestamps, highlight_words=highlight_words, progress=progress)
|
134 |
-
|
135 |
-
# Entry function for the full tab
|
136 |
-
def transcribe_webui_full(self, modelName, languageName, urlData, multipleFiles, microphoneData, task,
|
137 |
-
vad, vadMergeWindow, vadMaxMergeSize, vadPadding, vadPromptWindow, vadInitialPromptMode,
|
138 |
-
# Word timestamps
|
139 |
-
word_timestamps: bool, highlight_words: bool, prepend_punctuations: str, append_punctuations: str,
|
140 |
-
initial_prompt: str, temperature: float, best_of: int, beam_size: int, patience: float, length_penalty: float, suppress_tokens: str,
|
141 |
-
condition_on_previous_text: bool, fp16: bool, temperature_increment_on_fallback: float,
|
142 |
-
compression_ratio_threshold: float, logprob_threshold: float, no_speech_threshold: float,
|
143 |
-
diarization: bool = False, diarization_speakers: int = 2,
|
144 |
-
diarization_min_speakers = 1, diarization_max_speakers = 5):
|
145 |
-
|
146 |
-
return self.transcribe_webui_full_progress(modelName, languageName, urlData, multipleFiles, microphoneData, task,
|
147 |
-
vad, vadMergeWindow, vadMaxMergeSize, vadPadding, vadPromptWindow, vadInitialPromptMode,
|
148 |
-
word_timestamps, highlight_words, prepend_punctuations, append_punctuations,
|
149 |
-
initial_prompt, temperature, best_of, beam_size, patience, length_penalty, suppress_tokens,
|
150 |
-
condition_on_previous_text, fp16, temperature_increment_on_fallback,
|
151 |
-
compression_ratio_threshold, logprob_threshold, no_speech_threshold,
|
152 |
-
diarization, diarization_speakers,
|
153 |
-
diarization_min_speakers, diarization_max_speakers)
|
154 |
-
|
155 |
-
# Entry function for the full tab with progress
|
156 |
-
def transcribe_webui_full_progress(self, modelName, languageName, urlData, multipleFiles, microphoneData, task,
|
157 |
-
vad, vadMergeWindow, vadMaxMergeSize, vadPadding, vadPromptWindow, vadInitialPromptMode,
|
158 |
-
# Word timestamps
|
159 |
-
word_timestamps: bool, highlight_words: bool, prepend_punctuations: str, append_punctuations: str,
|
160 |
-
initial_prompt: str, temperature: float, best_of: int, beam_size: int, patience: float, length_penalty: float, suppress_tokens: str,
|
161 |
-
condition_on_previous_text: bool, fp16: bool, temperature_increment_on_fallback: float,
|
162 |
-
compression_ratio_threshold: float, logprob_threshold: float, no_speech_threshold: float,
|
163 |
-
diarization: bool = False, diarization_speakers: int = 2,
|
164 |
-
diarization_min_speakers = 1, diarization_max_speakers = 5,
|
165 |
-
progress=gr.Progress()):
|
166 |
-
|
167 |
-
# Handle temperature_increment_on_fallback
|
168 |
-
if temperature_increment_on_fallback is not None:
|
169 |
-
temperature = tuple(np.arange(temperature, 1.0 + 1e-6, temperature_increment_on_fallback))
|
170 |
-
else:
|
171 |
-
temperature = [temperature]
|
172 |
-
|
173 |
-
vadOptions = VadOptions(vad, vadMergeWindow, vadMaxMergeSize, vadPadding, vadPromptWindow, vadInitialPromptMode)
|
174 |
-
|
175 |
-
# Set diarization
|
176 |
-
if diarization:
|
177 |
-
self.set_diarization(auth_token=self.app_config.auth_token, num_speakers=diarization_speakers,
|
178 |
-
min_speakers=diarization_min_speakers, max_speakers=diarization_max_speakers)
|
179 |
-
else:
|
180 |
-
self.unset_diarization()
|
181 |
-
|
182 |
-
return self.transcribe_webui(modelName, languageName, urlData, multipleFiles, microphoneData, task, vadOptions,
|
183 |
-
initial_prompt=initial_prompt, temperature=temperature, best_of=best_of, beam_size=beam_size, patience=patience, length_penalty=length_penalty, suppress_tokens=suppress_tokens,
|
184 |
-
condition_on_previous_text=condition_on_previous_text, fp16=fp16,
|
185 |
-
compression_ratio_threshold=compression_ratio_threshold, logprob_threshold=logprob_threshold, no_speech_threshold=no_speech_threshold,
|
186 |
-
word_timestamps=word_timestamps, prepend_punctuations=prepend_punctuations, append_punctuations=append_punctuations, highlight_words=highlight_words,
|
187 |
-
progress=progress)
|
188 |
-
|
189 |
-
# Perform diarization given a specific input audio file and whisper file
|
190 |
-
def perform_extra(self, languageName, urlData, singleFile, whisper_file: str,
|
191 |
-
highlight_words: bool = False,
|
192 |
-
diarization: bool = False, diarization_speakers: int = 2, diarization_min_speakers = 1, diarization_max_speakers = 5, progress=gr.Progress()):
|
193 |
-
|
194 |
-
if whisper_file is None:
|
195 |
-
raise ValueError("whisper_file is required")
|
196 |
-
|
197 |
-
# Set diarization
|
198 |
-
if diarization:
|
199 |
-
self.set_diarization(auth_token=self.app_config.auth_token, num_speakers=diarization_speakers,
|
200 |
-
min_speakers=diarization_min_speakers, max_speakers=diarization_max_speakers)
|
201 |
-
else:
|
202 |
-
self.unset_diarization()
|
203 |
-
|
204 |
-
def custom_transcribe_file(source: AudioSource):
|
205 |
-
result = load_transcript(whisper_file.name)
|
206 |
-
|
207 |
-
# Set language if not set
|
208 |
-
if not "language" in result:
|
209 |
-
result["language"] = languageName
|
210 |
-
|
211 |
-
# Mark speakers
|
212 |
-
result = self._handle_diarization(source.source_path, result)
|
213 |
-
return result
|
214 |
-
|
215 |
-
multipleFiles = [singleFile] if singleFile else None
|
216 |
-
|
217 |
-
# Will return download, text, vtt
|
218 |
-
return self.transcribe_webui("base", "", urlData, multipleFiles, None, None, None,
|
219 |
-
progress=progress,highlight_words=highlight_words,
|
220 |
-
override_transcribe_file=custom_transcribe_file, override_max_sources=1)
|
221 |
-
|
222 |
-
def transcribe_webui(self, modelName, languageName, urlData, multipleFiles, microphoneData, task,
|
223 |
-
vadOptions: VadOptions, progress: gr.Progress = None, highlight_words: bool = False,
|
224 |
-
override_transcribe_file: Callable[[AudioSource], dict] = None, override_max_sources = None,
|
225 |
-
**decodeOptions: dict):
|
226 |
-
try:
|
227 |
-
sources = self.__get_source(urlData, multipleFiles, microphoneData)
|
228 |
-
|
229 |
-
if override_max_sources is not None and len(sources) > override_max_sources:
|
230 |
-
raise ValueError("Maximum number of sources is " + str(override_max_sources) + ", but " + str(len(sources)) + " were provided")
|
231 |
-
|
232 |
-
try:
|
233 |
-
selectedLanguage = languageName.lower() if len(languageName) > 0 else None
|
234 |
-
selectedModel = modelName if modelName is not None else "base"
|
235 |
-
|
236 |
-
if override_transcribe_file is None:
|
237 |
-
model = create_whisper_container(whisper_implementation=self.app_config.whisper_implementation,
|
238 |
-
model_name=selectedModel, compute_type=self.app_config.compute_type,
|
239 |
-
cache=self.model_cache, models=self.app_config.models)
|
240 |
-
else:
|
241 |
-
model = None
|
242 |
-
|
243 |
-
# Result
|
244 |
-
download = []
|
245 |
-
zip_file_lookup = {}
|
246 |
-
text = ""
|
247 |
-
vtt = ""
|
248 |
-
|
249 |
-
# Write result
|
250 |
-
downloadDirectory = tempfile.mkdtemp()
|
251 |
-
source_index = 0
|
252 |
-
|
253 |
-
outputDirectory = self.output_dir if self.output_dir is not None else downloadDirectory
|
254 |
-
|
255 |
-
# Progress
|
256 |
-
total_duration = sum([source.get_audio_duration() for source in sources])
|
257 |
-
current_progress = 0
|
258 |
-
|
259 |
-
# A listener that will report progress to Gradio
|
260 |
-
root_progress_listener = self._create_progress_listener(progress)
|
261 |
-
|
262 |
-
# Execute whisper
|
263 |
-
for source in sources:
|
264 |
-
source_prefix = ""
|
265 |
-
source_audio_duration = source.get_audio_duration()
|
266 |
-
|
267 |
-
if (len(sources) > 1):
|
268 |
-
# Prefix (minimum 2 digits)
|
269 |
-
source_index += 1
|
270 |
-
source_prefix = str(source_index).zfill(2) + "_"
|
271 |
-
print("Transcribing ", source.source_path)
|
272 |
-
|
273 |
-
scaled_progress_listener = SubTaskProgressListener(root_progress_listener,
|
274 |
-
base_task_total=total_duration,
|
275 |
-
sub_task_start=current_progress,
|
276 |
-
sub_task_total=source_audio_duration)
|
277 |
-
|
278 |
-
# Transcribe using the override function if specified
|
279 |
-
if override_transcribe_file is None:
|
280 |
-
result = self.transcribe_file(model, source.source_path, selectedLanguage, task, vadOptions, scaled_progress_listener, **decodeOptions)
|
281 |
-
else:
|
282 |
-
result = override_transcribe_file(source)
|
283 |
-
|
284 |
-
filePrefix = slugify(source_prefix + source.get_short_name(), allow_unicode=True)
|
285 |
-
|
286 |
-
# Update progress
|
287 |
-
current_progress += source_audio_duration
|
288 |
-
|
289 |
-
source_download, source_text, source_vtt = self.write_result(result, filePrefix, outputDirectory, highlight_words)
|
290 |
-
|
291 |
-
if len(sources) > 1:
|
292 |
-
# Add new line separators
|
293 |
-
if (len(source_text) > 0):
|
294 |
-
source_text += os.linesep + os.linesep
|
295 |
-
if (len(source_vtt) > 0):
|
296 |
-
source_vtt += os.linesep + os.linesep
|
297 |
-
|
298 |
-
# Append file name to source text too
|
299 |
-
source_text = source.get_full_name() + ":" + os.linesep + source_text
|
300 |
-
source_vtt = source.get_full_name() + ":" + os.linesep + source_vtt
|
301 |
-
|
302 |
-
# Add to result
|
303 |
-
download.extend(source_download)
|
304 |
-
text += source_text
|
305 |
-
vtt += source_vtt
|
306 |
-
|
307 |
-
if (len(sources) > 1):
|
308 |
-
# Zip files support at least 260 characters, but we'll play it safe and use 200
|
309 |
-
zipFilePrefix = slugify(source_prefix + source.get_short_name(max_length=200), allow_unicode=True)
|
310 |
-
|
311 |
-
# File names in ZIP file can be longer
|
312 |
-
for source_download_file in source_download:
|
313 |
-
# Get file postfix (after last -)
|
314 |
-
filePostfix = os.path.basename(source_download_file).split("-")[-1]
|
315 |
-
zip_file_name = zipFilePrefix + "-" + filePostfix
|
316 |
-
zip_file_lookup[source_download_file] = zip_file_name
|
317 |
-
|
318 |
-
# Create zip file from all sources
|
319 |
-
if len(sources) > 1:
|
320 |
-
downloadAllPath = os.path.join(downloadDirectory, "All_Output-" + datetime.now().strftime("%Y%m%d-%H%M%S") + ".zip")
|
321 |
-
|
322 |
-
with zipfile.ZipFile(downloadAllPath, 'w', zipfile.ZIP_DEFLATED) as zip:
|
323 |
-
for download_file in download:
|
324 |
-
# Get file name from lookup
|
325 |
-
zip_file_name = zip_file_lookup.get(download_file, os.path.basename(download_file))
|
326 |
-
zip.write(download_file, arcname=zip_file_name)
|
327 |
-
|
328 |
-
download.insert(0, downloadAllPath)
|
329 |
-
|
330 |
-
return download, text, vtt
|
331 |
-
|
332 |
-
finally:
|
333 |
-
# Cleanup source
|
334 |
-
if self.deleteUploadedFiles:
|
335 |
-
for source in sources:
|
336 |
-
print("Deleting source file " + source.source_path)
|
337 |
-
|
338 |
-
try:
|
339 |
-
os.remove(source.source_path)
|
340 |
-
except Exception as e:
|
341 |
-
# Ignore error - it's just a cleanup
|
342 |
-
print("Error deleting source file " + source.source_path + ": " + str(e))
|
343 |
-
|
344 |
-
except ExceededMaximumDuration as e:
|
345 |
-
return [], ("[ERROR]: Maximum remote video length is " + str(e.maxDuration) + "s, file was " + str(e.videoDuration) + "s"), "[ERROR]"
|
346 |
-
|
347 |
-
def transcribe_file(self, model: AbstractWhisperContainer, audio_path: str, language: str, task: str = None,
|
348 |
-
vadOptions: VadOptions = VadOptions(),
|
349 |
-
progressListener: ProgressListener = None, **decodeOptions: dict):
|
350 |
-
|
351 |
-
initial_prompt = decodeOptions.pop('initial_prompt', None)
|
352 |
-
|
353 |
-
if progressListener is None:
|
354 |
-
# Default progress listener
|
355 |
-
progressListener = ProgressListener()
|
356 |
-
|
357 |
-
if ('task' in decodeOptions):
|
358 |
-
task = decodeOptions.pop('task')
|
359 |
-
|
360 |
-
initial_prompt_mode = vadOptions.vadInitialPromptMode
|
361 |
-
|
362 |
-
# Set default initial prompt mode
|
363 |
-
if (initial_prompt_mode is None):
|
364 |
-
initial_prompt_mode = VadInitialPromptMode.PREPREND_FIRST_SEGMENT
|
365 |
-
|
366 |
-
if (initial_prompt_mode == VadInitialPromptMode.PREPEND_ALL_SEGMENTS or
|
367 |
-
initial_prompt_mode == VadInitialPromptMode.PREPREND_FIRST_SEGMENT):
|
368 |
-
# Prepend initial prompt
|
369 |
-
prompt_strategy = PrependPromptStrategy(initial_prompt, initial_prompt_mode)
|
370 |
-
elif (vadOptions.vadInitialPromptMode == VadInitialPromptMode.JSON_PROMPT_MODE):
|
371 |
-
# Use a JSON format to specify the prompt for each segment
|
372 |
-
prompt_strategy = JsonPromptStrategy(initial_prompt)
|
373 |
-
else:
|
374 |
-
raise ValueError("Invalid vadInitialPromptMode: " + initial_prompt_mode)
|
375 |
-
|
376 |
-
# Callable for processing an audio file
|
377 |
-
whisperCallable = model.create_callback(language, task, prompt_strategy=prompt_strategy, **decodeOptions)
|
378 |
-
|
379 |
-
# The results
|
380 |
-
if (vadOptions.vad == 'silero-vad'):
|
381 |
-
# Silero VAD where non-speech gaps are transcribed
|
382 |
-
process_gaps = self._create_silero_config(NonSpeechStrategy.CREATE_SEGMENT, vadOptions)
|
383 |
-
result = self.process_vad(audio_path, whisperCallable, self.vad_model, process_gaps, progressListener=progressListener)
|
384 |
-
elif (vadOptions.vad == 'silero-vad-skip-gaps'):
|
385 |
-
# Silero VAD where non-speech gaps are simply ignored
|
386 |
-
skip_gaps = self._create_silero_config(NonSpeechStrategy.SKIP, vadOptions)
|
387 |
-
result = self.process_vad(audio_path, whisperCallable, self.vad_model, skip_gaps, progressListener=progressListener)
|
388 |
-
elif (vadOptions.vad == 'silero-vad-expand-into-gaps'):
|
389 |
-
# Use Silero VAD where speech-segments are expanded into non-speech gaps
|
390 |
-
expand_gaps = self._create_silero_config(NonSpeechStrategy.EXPAND_SEGMENT, vadOptions)
|
391 |
-
result = self.process_vad(audio_path, whisperCallable, self.vad_model, expand_gaps, progressListener=progressListener)
|
392 |
-
elif (vadOptions.vad == 'periodic-vad'):
|
393 |
-
# Very simple VAD - mark every 5 minutes as speech. This makes it less likely that Whisper enters an infinite loop, but
|
394 |
-
# it may create a break in the middle of a sentence, causing some artifacts.
|
395 |
-
periodic_vad = VadPeriodicTranscription()
|
396 |
-
period_config = PeriodicTranscriptionConfig(periodic_duration=vadOptions.vadMaxMergeSize, max_prompt_window=vadOptions.vadPromptWindow)
|
397 |
-
result = self.process_vad(audio_path, whisperCallable, periodic_vad, period_config, progressListener=progressListener)
|
398 |
-
|
399 |
-
else:
|
400 |
-
if (self._has_parallel_devices()):
|
401 |
-
# Use a simple period transcription instead, as we need to use the parallel context
|
402 |
-
periodic_vad = VadPeriodicTranscription()
|
403 |
-
period_config = PeriodicTranscriptionConfig(periodic_duration=math.inf, max_prompt_window=1)
|
404 |
-
|
405 |
-
result = self.process_vad(audio_path, whisperCallable, periodic_vad, period_config, progressListener=progressListener)
|
406 |
-
else:
|
407 |
-
# Default VAD
|
408 |
-
result = whisperCallable.invoke(audio_path, 0, None, None, progress_listener=progressListener)
|
409 |
-
|
410 |
-
# Diarization
|
411 |
-
result = self._handle_diarization(audio_path, result)
|
412 |
-
return result
|
413 |
-
|
414 |
-
def _handle_diarization(self, audio_path: str, input: dict):
|
415 |
-
if self.diarization and self.diarization_kwargs:
|
416 |
-
print("Diarizing ", audio_path)
|
417 |
-
diarization_result = list(self.diarization.run(audio_path, **self.diarization_kwargs))
|
418 |
-
|
419 |
-
# Print result
|
420 |
-
print("Diarization result: ")
|
421 |
-
for entry in diarization_result:
|
422 |
-
print(f" start={entry.start:.1f}s stop={entry.end:.1f}s speaker_{entry.speaker}")
|
423 |
-
|
424 |
-
# Add speakers to result
|
425 |
-
input = self.diarization.mark_speakers(diarization_result, input)
|
426 |
-
|
427 |
-
return input
|
428 |
-
|
429 |
-
def _create_progress_listener(self, progress: gr.Progress):
|
430 |
-
if (progress is None):
|
431 |
-
# Dummy progress listener
|
432 |
-
return ProgressListener()
|
433 |
-
|
434 |
-
class ForwardingProgressListener(ProgressListener):
|
435 |
-
def __init__(self, progress: gr.Progress):
|
436 |
-
self.progress = progress
|
437 |
-
|
438 |
-
def on_progress(self, current: Union[int, float], total: Union[int, float]):
|
439 |
-
# From 0 to 1
|
440 |
-
self.progress(current / total)
|
441 |
-
|
442 |
-
def on_finished(self):
|
443 |
-
self.progress(1)
|
444 |
-
|
445 |
-
return ForwardingProgressListener(progress)
|
446 |
-
|
447 |
-
def process_vad(self, audio_path, whisperCallable, vadModel: AbstractTranscription, vadConfig: TranscriptionConfig,
|
448 |
-
progressListener: ProgressListener = None):
|
449 |
-
if (not self._has_parallel_devices()):
|
450 |
-
# No parallel devices, so just run the VAD and Whisper in sequence
|
451 |
-
return vadModel.transcribe(audio_path, whisperCallable, vadConfig, progressListener=progressListener)
|
452 |
-
|
453 |
-
gpu_devices = self.parallel_device_list
|
454 |
-
|
455 |
-
if (gpu_devices is None or len(gpu_devices) == 0):
|
456 |
-
# No GPU devices specified, pass the current environment variable to the first GPU process. This may be NULL.
|
457 |
-
gpu_devices = [os.environ.get("CUDA_VISIBLE_DEVICES", None)]
|
458 |
-
|
459 |
-
# Create parallel context if needed
|
460 |
-
if (self.gpu_parallel_context is None):
|
461 |
-
# Create a context wih processes and automatically clear the pool after 1 hour of inactivity
|
462 |
-
self.gpu_parallel_context = ParallelContext(num_processes=len(gpu_devices), auto_cleanup_timeout_seconds=self.vad_process_timeout)
|
463 |
-
# We also need a CPU context for the VAD
|
464 |
-
if (self.cpu_parallel_context is None):
|
465 |
-
self.cpu_parallel_context = ParallelContext(num_processes=self.vad_cpu_cores, auto_cleanup_timeout_seconds=self.vad_process_timeout)
|
466 |
-
|
467 |
-
parallel_vad = ParallelTranscription()
|
468 |
-
return parallel_vad.transcribe_parallel(transcription=vadModel, audio=audio_path, whisperCallable=whisperCallable,
|
469 |
-
config=vadConfig, cpu_device_count=self.vad_cpu_cores, gpu_devices=gpu_devices,
|
470 |
-
cpu_parallel_context=self.cpu_parallel_context, gpu_parallel_context=self.gpu_parallel_context,
|
471 |
-
progress_listener=progressListener)
|
472 |
-
|
473 |
-
def _has_parallel_devices(self):
|
474 |
-
return (self.parallel_device_list is not None and len(self.parallel_device_list) > 0) or self.vad_cpu_cores > 1
|
475 |
-
|
476 |
-
def _concat_prompt(self, prompt1, prompt2):
|
477 |
-
if (prompt1 is None):
|
478 |
-
return prompt2
|
479 |
-
elif (prompt2 is None):
|
480 |
-
return prompt1
|
481 |
-
else:
|
482 |
-
return prompt1 + " " + prompt2
|
483 |
-
|
484 |
-
def _create_silero_config(self, non_speech_strategy: NonSpeechStrategy, vadOptions: VadOptions):
|
485 |
-
# Use Silero VAD
|
486 |
-
if (self.vad_model is None):
|
487 |
-
self.vad_model = VadSileroTranscription()
|
488 |
-
|
489 |
-
config = TranscriptionConfig(non_speech_strategy = non_speech_strategy,
|
490 |
-
max_silent_period=vadOptions.vadMergeWindow, max_merge_size=vadOptions.vadMaxMergeSize,
|
491 |
-
segment_padding_left=vadOptions.vadPadding, segment_padding_right=vadOptions.vadPadding,
|
492 |
-
max_prompt_window=vadOptions.vadPromptWindow)
|
493 |
-
|
494 |
-
return config
|
495 |
-
|
496 |
-
def write_result(self, result: dict, source_name: str, output_dir: str, highlight_words: bool = False):
|
497 |
-
if not os.path.exists(output_dir):
|
498 |
-
os.makedirs(output_dir)
|
499 |
-
|
500 |
-
text = result["text"]
|
501 |
-
language = result["language"] if "language" in result else None
|
502 |
-
languageMaxLineWidth = self.__get_max_line_width(language)
|
503 |
-
|
504 |
-
# We always create the JSON file for debugging purposes
|
505 |
-
json_result = json.dumps(result, indent=4, ensure_ascii=False)
|
506 |
-
json_file = self.__create_file(json_result, output_dir, source_name + "-result.json")
|
507 |
-
print("Created JSON file " + json_file)
|
508 |
-
|
509 |
-
print("Max line width " + str(languageMaxLineWidth))
|
510 |
-
vtt = self.__get_subs(result["segments"], "vtt", languageMaxLineWidth, highlight_words=highlight_words)
|
511 |
-
srt = self.__get_subs(result["segments"], "srt", languageMaxLineWidth, highlight_words=highlight_words)
|
512 |
-
|
513 |
-
output_files = []
|
514 |
-
output_files.append(self.__create_file(srt, output_dir, source_name + "-subs.srt"));
|
515 |
-
output_files.append(self.__create_file(vtt, output_dir, source_name + "-subs.vtt"));
|
516 |
-
output_files.append(self.__create_file(text, output_dir, source_name + "-transcript.txt"));
|
517 |
-
output_files.append(json_file)
|
518 |
-
|
519 |
-
return output_files, text, vtt
|
520 |
-
|
521 |
-
def clear_cache(self):
|
522 |
-
self.model_cache.clear()
|
523 |
-
self.vad_model = None
|
524 |
-
|
525 |
-
def __get_source(self, urlData, multipleFiles, microphoneData):
|
526 |
-
return get_audio_source_collection(urlData, multipleFiles, microphoneData, self.inputAudioMaxDuration)
|
527 |
-
|
528 |
-
def __get_max_line_width(self, language: str) -> int:
|
529 |
-
if (language and language.lower() in ["japanese", "ja", "chinese", "zh"]):
|
530 |
-
# Chinese characters and kana are wider, so limit line length to 40 characters
|
531 |
-
return 40
|
532 |
-
else:
|
533 |
-
# TODO: Add more languages
|
534 |
-
# 80 latin characters should fit on a 1080p/720p screen
|
535 |
-
return 80
|
536 |
-
|
537 |
-
def __get_subs(self, segments: Iterator[dict], format: str, maxLineWidth: int, highlight_words: bool = False) -> str:
|
538 |
-
segmentStream = StringIO()
|
539 |
-
|
540 |
-
if format == 'vtt':
|
541 |
-
write_vtt(segments, file=segmentStream, maxLineWidth=maxLineWidth, highlight_words=highlight_words)
|
542 |
-
elif format == 'srt':
|
543 |
-
write_srt(segments, file=segmentStream, maxLineWidth=maxLineWidth, highlight_words=highlight_words)
|
544 |
-
else:
|
545 |
-
raise Exception("Unknown format " + format)
|
546 |
-
|
547 |
-
segmentStream.seek(0)
|
548 |
-
return segmentStream.read()
|
549 |
-
|
550 |
-
def __create_file(self, text: str, directory: str, fileName: str) -> str:
|
551 |
-
# Write the text to a file
|
552 |
-
with open(os.path.join(directory, fileName), 'w+', encoding="utf-8") as file:
|
553 |
-
file.write(text)
|
554 |
-
|
555 |
-
return file.name
|
556 |
-
|
557 |
-
def close(self):
|
558 |
-
print("Closing parallel contexts")
|
559 |
-
self.clear_cache()
|
560 |
-
|
561 |
-
if (self.gpu_parallel_context is not None):
|
562 |
-
self.gpu_parallel_context.close()
|
563 |
-
if (self.cpu_parallel_context is not None):
|
564 |
-
self.cpu_parallel_context.close()
|
565 |
-
|
566 |
-
# Cleanup diarization
|
567 |
-
if (self.diarization is not None):
|
568 |
-
self.diarization.cleanup()
|
569 |
-
self.diarization = None
|
570 |
-
|
571 |
-
def create_ui(app_config: ApplicationConfig):
|
572 |
-
ui = WhisperTranscriber(app_config.input_audio_max_duration, app_config.vad_process_timeout, app_config.vad_cpu_cores,
|
573 |
-
app_config.delete_uploaded_files, app_config.output_dir, app_config)
|
574 |
-
|
575 |
-
# Specify a list of devices to use for parallel processing
|
576 |
-
ui.set_parallel_devices(app_config.vad_parallel_devices)
|
577 |
-
ui.set_auto_parallel(app_config.auto_parallel)
|
578 |
-
|
579 |
-
is_whisper = False
|
580 |
-
|
581 |
-
if app_config.whisper_implementation == "whisper":
|
582 |
-
implementation_name = "Whisper"
|
583 |
-
is_whisper = True
|
584 |
-
elif app_config.whisper_implementation in ["faster-whisper", "faster_whisper"]:
|
585 |
-
implementation_name = "Faster Whisper"
|
586 |
-
else:
|
587 |
-
# Try to convert from camel-case to title-case
|
588 |
-
implementation_name = app_config.whisper_implementation.title().replace("_", " ").replace("-", " ")
|
589 |
-
|
590 |
-
ui_description = implementation_name + " is a general-purpose speech recognition model. It is trained on a large dataset of diverse "
|
591 |
-
ui_description += " audio and is also a multi-task model that can perform multilingual speech recognition "
|
592 |
-
ui_description += " as well as speech translation and language identification. "
|
593 |
-
|
594 |
-
ui_description += "\n\n\n\nFor longer audio files (>10 minutes) not in English, it is recommended that you select Silero VAD (Voice Activity Detector) in the VAD option."
|
595 |
-
|
596 |
-
# Recommend faster-whisper
|
597 |
-
if is_whisper:
|
598 |
-
ui_description += "\n\n\n\nFor faster inference on GPU, try [faster-whisper](https://huggingface.co/spaces/aadnk/faster-whisper-webui)."
|
599 |
-
|
600 |
-
if app_config.input_audio_max_duration > 0:
|
601 |
-
ui_description += "\n\n" + "Max audio file length: " + str(app_config.input_audio_max_duration) + " s"
|
602 |
-
|
603 |
-
ui_article = "Read the [documentation here](https://gitlab.com/aadnk/whisper-webui/-/blob/main/docs/options.md)."
|
604 |
-
|
605 |
-
whisper_models = app_config.get_model_names()
|
606 |
-
|
607 |
-
common_inputs = lambda : [
|
608 |
-
gr.Dropdown(choices=whisper_models, value=app_config.default_model_name, label="Model"),
|
609 |
-
gr.Dropdown(choices=sorted(get_language_names()), label="Language", value=app_config.language),
|
610 |
-
gr.Text(label="URL (YouTube, etc.)"),
|
611 |
-
gr.File(label="Upload Files", file_count="multiple"),
|
612 |
-
gr.Audio(source="microphone", type="filepath", label="Microphone Input"),
|
613 |
-
gr.Dropdown(choices=["transcribe", "translate"], label="Task", value=app_config.task),
|
614 |
-
]
|
615 |
-
|
616 |
-
common_vad_inputs = lambda : [
|
617 |
-
gr.Dropdown(choices=["none", "silero-vad", "silero-vad-skip-gaps", "silero-vad-expand-into-gaps", "periodic-vad"], value=app_config.default_vad, label="VAD"),
|
618 |
-
gr.Number(label="VAD - Merge Window (s)", precision=0, value=app_config.vad_merge_window),
|
619 |
-
gr.Number(label="VAD - Max Merge Size (s)", precision=0, value=app_config.vad_max_merge_size),
|
620 |
-
]
|
621 |
|
622 |
-
|
623 |
-
|
624 |
-
|
625 |
-
|
626 |
-
|
627 |
-
has_diarization_libs = Diarization.has_libraries()
|
628 |
-
|
629 |
-
if not has_diarization_libs:
|
630 |
-
print("Diarization libraries not found - disabling diarization")
|
631 |
-
app_config.diarization = False
|
632 |
-
|
633 |
-
common_diarization_inputs = lambda : [
|
634 |
-
gr.Checkbox(label="Diarization", value=app_config.diarization, interactive=has_diarization_libs),
|
635 |
-
gr.Number(label="Diarization - Speakers", precision=0, value=app_config.diarization_speakers, interactive=has_diarization_libs)
|
636 |
-
]
|
637 |
-
|
638 |
-
is_queue_mode = app_config.queue_concurrency_count is not None and app_config.queue_concurrency_count > 0
|
639 |
-
|
640 |
-
simple_transcribe = gr.Interface(fn=ui.transcribe_webui_simple_progress if is_queue_mode else ui.transcribe_webui_simple,
|
641 |
-
description=ui_description, article=ui_article, inputs=[
|
642 |
-
*common_inputs(),
|
643 |
-
*common_vad_inputs(),
|
644 |
-
*common_word_timestamps_inputs(),
|
645 |
-
*common_diarization_inputs(),
|
646 |
-
], outputs=[
|
647 |
-
gr.File(label="Download"),
|
648 |
-
gr.Text(label="Transcription"),
|
649 |
-
gr.Text(label="Segments")
|
650 |
-
])
|
651 |
-
|
652 |
-
full_description = ui_description + "\n\n\n\n" + "Be careful when changing some of the options in the full interface - this can cause the model to crash."
|
653 |
-
|
654 |
-
full_transcribe = gr.Interface(fn=ui.transcribe_webui_full_progress if is_queue_mode else ui.transcribe_webui_full,
|
655 |
-
description=full_description, article=ui_article, inputs=[
|
656 |
-
*common_inputs(),
|
657 |
|
658 |
-
|
659 |
-
|
660 |
-
|
661 |
-
|
662 |
-
|
663 |
-
|
664 |
-
|
665 |
-
|
666 |
|
667 |
-
|
668 |
-
|
669 |
-
|
670 |
-
|
671 |
-
gr.Number(label="Patience - Zero temperature", value=app_config.patience),
|
672 |
-
gr.Number(label="Length Penalty - Any temperature", value=app_config.length_penalty),
|
673 |
-
gr.Text(label="Suppress Tokens - Comma-separated list of token IDs", value=app_config.suppress_tokens),
|
674 |
-
gr.Checkbox(label="Condition on previous text", value=app_config.condition_on_previous_text),
|
675 |
-
gr.Checkbox(label="FP16", value=app_config.fp16),
|
676 |
-
gr.Number(label="Temperature increment on fallback", value=app_config.temperature_increment_on_fallback),
|
677 |
-
gr.Number(label="Compression ratio threshold", value=app_config.compression_ratio_threshold),
|
678 |
-
gr.Number(label="Logprob threshold", value=app_config.logprob_threshold),
|
679 |
-
gr.Number(label="No speech threshold", value=app_config.no_speech_threshold),
|
680 |
|
681 |
-
|
682 |
-
|
683 |
-
gr.Number(label="Diarization - Max Speakers", precision=0, value=app_config.diarization_max_speakers, interactive=has_diarization_libs),
|
684 |
|
685 |
-
|
686 |
-
|
687 |
-
|
688 |
-
gr.Text(label="Segments")
|
689 |
-
])
|
690 |
|
691 |
-
|
692 |
-
|
693 |
-
|
694 |
-
gr.Text(label="URL (YouTube, etc.)"),
|
695 |
-
gr.File(label="Upload Audio File", file_count="single"),
|
696 |
-
gr.File(label="Upload JSON/SRT File", file_count="single"),
|
697 |
-
gr.Checkbox(label="Word Timestamps - Highlight Words", value=app_config.highlight_words),
|
698 |
|
699 |
-
|
700 |
-
gr.Number(label="Diarization - Min Speakers", precision=0, value=app_config.diarization_min_speakers, interactive=has_diarization_libs),
|
701 |
-
gr.Number(label="Diarization - Max Speakers", precision=0, value=app_config.diarization_max_speakers, interactive=has_diarization_libs),
|
702 |
-
|
703 |
-
], outputs=[
|
704 |
-
gr.File(label="Download"),
|
705 |
-
gr.Text(label="Transcription"),
|
706 |
-
gr.Text(label="Segments")
|
707 |
-
])
|
708 |
-
|
709 |
-
demo = gr.TabbedInterface([simple_transcribe, full_transcribe, perform_extra_interface], tab_names=["Simple", "Full", "Extra"])
|
710 |
-
|
711 |
-
# Queue up the demo
|
712 |
-
if is_queue_mode:
|
713 |
-
demo.queue(concurrency_count=app_config.queue_concurrency_count)
|
714 |
-
print("Queue mode enabled (concurrency count: " + str(app_config.queue_concurrency_count) + ")")
|
715 |
-
else:
|
716 |
-
print("Queue mode disabled - progress bars will not be shown.")
|
717 |
-
|
718 |
-
demo.launch(share=app_config.share, server_name=app_config.server_name, server_port=app_config.server_port)
|
719 |
-
|
720 |
-
# Clean up
|
721 |
-
ui.close()
|
722 |
-
|
723 |
-
if __name__ == '__main__':
|
724 |
-
default_app_config = ApplicationConfig.create_default()
|
725 |
-
whisper_models = default_app_config.get_model_names()
|
726 |
-
|
727 |
-
# Environment variable overrides
|
728 |
-
default_whisper_implementation = os.environ.get("WHISPER_IMPLEMENTATION", default_app_config.whisper_implementation)
|
729 |
-
|
730 |
-
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
|
731 |
-
parser.add_argument("--input_audio_max_duration", type=int, default=default_app_config.input_audio_max_duration, \
|
732 |
-
help="Maximum audio file length in seconds, or -1 for no limit.") # 600
|
733 |
-
parser.add_argument("--share", type=bool, default=default_app_config.share, \
|
734 |
-
help="True to share the app on HuggingFace.") # False
|
735 |
-
parser.add_argument("--server_name", type=str, default=default_app_config.server_name, \
|
736 |
-
help="The host or IP to bind to. If None, bind to localhost.") # None
|
737 |
-
parser.add_argument("--server_port", type=int, default=default_app_config.server_port, \
|
738 |
-
help="The port to bind to.") # 7860
|
739 |
-
parser.add_argument("--queue_concurrency_count", type=int, default=default_app_config.queue_concurrency_count, \
|
740 |
-
help="The number of concurrent requests to process.") # 1
|
741 |
-
parser.add_argument("--default_model_name", type=str, choices=whisper_models, default=default_app_config.default_model_name, \
|
742 |
-
help="The default model name.") # medium
|
743 |
-
parser.add_argument("--default_vad", type=str, default=default_app_config.default_vad, \
|
744 |
-
help="The default VAD.") # silero-vad
|
745 |
-
parser.add_argument("--vad_initial_prompt_mode", type=str, default=default_app_config.vad_initial_prompt_mode, choices=VAD_INITIAL_PROMPT_MODE_VALUES, \
|
746 |
-
help="Whether or not to prepend the initial prompt to each VAD segment (prepend_all_segments), or just the first segment (prepend_first_segment)") # prepend_first_segment
|
747 |
-
parser.add_argument("--vad_parallel_devices", type=str, default=default_app_config.vad_parallel_devices, \
|
748 |
-
help="A commma delimited list of CUDA devices to use for parallel processing. If None, disable parallel processing.") # ""
|
749 |
-
parser.add_argument("--vad_cpu_cores", type=int, default=default_app_config.vad_cpu_cores, \
|
750 |
-
help="The number of CPU cores to use for VAD pre-processing.") # 1
|
751 |
-
parser.add_argument("--vad_process_timeout", type=float, default=default_app_config.vad_process_timeout, \
|
752 |
-
help="The number of seconds before inactivate processes are terminated. Use 0 to close processes immediately, or None for no timeout.") # 1800
|
753 |
-
parser.add_argument("--auto_parallel", type=bool, default=default_app_config.auto_parallel, \
|
754 |
-
help="True to use all available GPUs and CPU cores for processing. Use vad_cpu_cores/vad_parallel_devices to specify the number of CPU cores/GPUs to use.") # False
|
755 |
-
parser.add_argument("--output_dir", "-o", type=str, default=default_app_config.output_dir, \
|
756 |
-
help="directory to save the outputs")
|
757 |
-
parser.add_argument("--whisper_implementation", type=str, default=default_whisper_implementation, choices=["whisper", "faster-whisper"],\
|
758 |
-
help="the Whisper implementation to use")
|
759 |
-
parser.add_argument("--compute_type", type=str, default=default_app_config.compute_type, choices=["default", "auto", "int8", "int8_float16", "int16", "float16", "float32"], \
|
760 |
-
help="the compute type to use for inference")
|
761 |
-
parser.add_argument("--threads", type=optional_int, default=0,
|
762 |
-
help="number of threads used by torch for CPU inference; supercedes MKL_NUM_THREADS/OMP_NUM_THREADS")
|
763 |
-
|
764 |
-
parser.add_argument('--auth_token', type=str, default=default_app_config.auth_token, help='HuggingFace API Token (optional)')
|
765 |
-
parser.add_argument("--diarization", type=str2bool, default=default_app_config.diarization, \
|
766 |
-
help="whether to perform speaker diarization")
|
767 |
-
parser.add_argument("--diarization_num_speakers", type=int, default=default_app_config.diarization_speakers, help="Number of speakers")
|
768 |
-
parser.add_argument("--diarization_min_speakers", type=int, default=default_app_config.diarization_min_speakers, help="Minimum number of speakers")
|
769 |
-
parser.add_argument("--diarization_max_speakers", type=int, default=default_app_config.diarization_max_speakers, help="Maximum number of speakers")
|
770 |
-
parser.add_argument("--diarization_process_timeout", type=int, default=default_app_config.diarization_process_timeout, \
|
771 |
-
help="Number of seconds before inactivate diarization processes are terminated. Use 0 to close processes immediately, or None for no timeout.")
|
772 |
|
773 |
-
|
|
|
774 |
|
775 |
-
|
|
|
|
|
776 |
|
777 |
-
|
778 |
-
|
779 |
|
780 |
-
|
781 |
-
create_ui(app_config=updated_config)
|
|
|
1 |
+
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
import os
|
|
|
3 |
import tempfile
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
import ffmpeg
|
5 |
+
import json
|
6 |
+
from huggingface_hub import InferenceApi
|
7 |
+
from typing import List, Dict, Tuple
|
8 |
+
|
9 |
+
# 🔹 Constants
|
10 |
+
MODEL_NAME: str = "ivrit-ai/faster-whisper-v2-d4"
|
11 |
+
TRANSLATION_MODEL_NAME: str = "dicta-il/dictalm2.0-GGUF"
|
12 |
+
TEMP_DIR: str = tempfile.gettempdir()
|
13 |
+
|
14 |
+
# 🔹 Load Hugging Face Inference API
|
15 |
+
ASR_API = InferenceApi(repo_id=MODEL_NAME)
|
16 |
+
TRANSLATION_API = InferenceApi(repo_id=TRANSLATION_MODEL_NAME)
|
17 |
+
|
18 |
+
def convert_audio(audio_path: str) -> str:
|
19 |
+
"""Converts an audio file to 16kHz WAV format for compatibility."""
|
20 |
+
converted_path = os.path.join(TEMP_DIR, "converted.wav")
|
21 |
+
(
|
22 |
+
ffmpeg
|
23 |
+
.input(audio_path)
|
24 |
+
.output(converted_path, format="wav", ar="16000")
|
25 |
+
.run(overwrite_output=True, quiet=True)
|
26 |
+
)
|
27 |
+
return converted_path
|
28 |
+
|
29 |
+
def transcribe_audio(file: str, translate: bool) -> Tuple[str, str]:
|
30 |
+
"""Transcribes audio and optionally translates it using Hugging Face API."""
|
31 |
+
audio_path = file if file.endswith(".wav") else convert_audio(file)
|
32 |
+
|
33 |
+
with open(audio_path, "rb") as audio_file:
|
34 |
+
result = ASR_API(inputs=audio_file)
|
35 |
+
|
36 |
+
segments = result.get("segments", [])
|
37 |
+
subtitles: List[Dict[str, str]] = []
|
38 |
+
transcribed_text: str = ""
|
39 |
+
|
40 |
+
for segment in segments:
|
41 |
+
hebrew_text = segment["text"]
|
42 |
+
start_time = segment["start"]
|
43 |
+
end_time = segment["end"]
|
44 |
+
eng_translation = ""
|
45 |
+
|
46 |
+
if translate:
|
47 |
+
eng_translation = TRANSLATION_API(inputs=hebrew_text)[0]["translation_text"]
|
48 |
+
|
49 |
+
subtitles.append({
|
50 |
+
"start": start_time,
|
51 |
+
"end": end_time,
|
52 |
+
"text": hebrew_text,
|
53 |
+
"translation": eng_translation if translate else None
|
54 |
+
})
|
55 |
+
|
56 |
+
transcribed_text += f"{hebrew_text} "
|
57 |
+
|
58 |
+
return json.dumps(subtitles), transcribed_text
|
59 |
+
|
60 |
+
# 🔹 Inject WebGPU-compatible JavaScript via `gr.HTML()`
|
61 |
+
webgpu_script = """
|
62 |
+
<script type="module">
|
63 |
+
import { pipeline } from 'https://cdn.jsdelivr.net/npm/@xenova/transformers@latest';
|
64 |
+
|
65 |
+
let asr;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
|
67 |
+
async function loadModel() {
|
68 |
+
asr = await pipeline("automatic-speech-recognition", "openai/whisper-large-v3");
|
69 |
+
console.log("WebGPU ASR model loaded.");
|
70 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
|
72 |
+
async function transcribe(audioFile) {
|
73 |
+
if (!asr) {
|
74 |
+
console.error("Model not loaded.");
|
75 |
+
return;
|
76 |
+
}
|
77 |
+
const result = await asr(audioFile);
|
78 |
+
document.getElementById("output").innerText = result.text;
|
79 |
+
}
|
80 |
|
81 |
+
document.getElementById("upload").addEventListener("change", async (event) => {
|
82 |
+
const file = event.target.files[0];
|
83 |
+
transcribe(file);
|
84 |
+
});
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
|
86 |
+
loadModel();
|
87 |
+
</script>
|
|
|
88 |
|
89 |
+
<input type="file" id="upload" accept="audio/*">
|
90 |
+
<p id="output">Transcription will appear here.</p>
|
91 |
+
"""
|
|
|
|
|
92 |
|
93 |
+
# 🔹 Gradio UI
|
94 |
+
with gr.Blocks() as demo:
|
95 |
+
gr.Markdown("# WhatShutup: Transcribe WhatsApp Voice Messages with WebGPU Support")
|
|
|
|
|
|
|
|
|
96 |
|
97 |
+
webgpu_component = gr.HTML(webgpu_script)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
|
99 |
+
audio_input = gr.Audio(source="upload", type="filepath", label="Upload Audio File")
|
100 |
+
translate_checkbox = gr.Checkbox(label="Translate to English?", value=False)
|
101 |
|
102 |
+
with gr.Row():
|
103 |
+
audio_player = gr.Audio(source="upload", type="filepath", label="Playback")
|
104 |
+
transcript_output = gr.Textbox(label="Transcription & Subtitles", lines=10)
|
105 |
|
106 |
+
submit_btn = gr.Button("Transcribe")
|
107 |
+
submit_btn.click(transcribe_audio, inputs=[audio_input, translate_checkbox], outputs=[audio_player, transcript_output])
|
108 |
|
109 |
+
demo.launch()
|
|