Spaces:
Running
Running
File size: 58,260 Bytes
40b3f9e e35aecd 40b3f9e f96fc4d 40b3f9e 7c1237e 1c18b3d d9fe1f1 f8b99f0 635fb63 40b3f9e 7add5f8 8d5a056 7add5f8 8d5a056 7add5f8 40b3f9e 7add5f8 40b3f9e 0c13f05 40b3f9e 1f8af3b 7add5f8 1f8af3b 40b3f9e 7add5f8 40b3f9e a83dd80 40b3f9e 7add5f8 40b3f9e 7add5f8 40b3f9e 7add5f8 40b3f9e 7add5f8 40b3f9e 7add5f8 40b3f9e 7add5f8 a4a5578 3898559 40b3f9e 7add5f8 3898559 7add5f8 40b3f9e 7add5f8 40b3f9e f99e269 40b3f9e f99e269 40b3f9e f99e269 40b3f9e 35efadd 40b3f9e 35efadd 40b3f9e 35efadd 40b3f9e e2e15a2 40b3f9e e2e15a2 40b3f9e 526a5f6 40b3f9e 0489670 e2e15a2 40b3f9e 526a5f6 e2e15a2 40b3f9e e2e15a2 40b3f9e e2e15a2 526a5f6 a332ef9 df8571d a332ef9 df8571d a332ef9 df8571d 40b3f9e e2e15a2 40b3f9e e2e15a2 40b3f9e a12a54f 40b3f9e 940ca8e 4e27cf9 40b3f9e a12a54f 40b3f9e 8953421 40b3f9e 4e27cf9 40b3f9e 4e27cf9 40b3f9e 940ca8e 40b3f9e 4e27cf9 40b3f9e f99e269 c6f940f f99e269 c6f940f f99e269 c6f940f f99e269 c6f940f f99e269 38d8739 f99e269 1320e5b f99e269 1320e5b f99e269 1320e5b f99e269 1320e5b f99e269 c34772c f99e269 d9fe1f1 f99e269 d9fe1f1 f99e269 f57819a d9fe1f1 f99e269 f57819a f99e269 7add5f8 f99e269 f57819a bbc39f5 f99e269 bbc39f5 f57819a bbc39f5 f57819a 7add5f8 f57819a bbc39f5 f57819a bbc39f5 f57819a bbc39f5 f57819a f99e269 f57819a f99e269 f57819a bbc39f5 f57819a f99e269 f57819a f99e269 7add5f8 f57819a f99e269 f57819a f99e269 f57819a f99e269 66933cf 7add5f8 d9fe1f1 b83a564 d9fe1f1 c34772c 7d826f9 d9fe1f1 7add5f8 dd10881 7add5f8 1c18b3d 7add5f8 7d826f9 40b3f9e e35aecd 40b3f9e a83dd80 40b3f9e a82093d 40b3f9e a82093d 40b3f9e 129e2e0 40b3f9e 129e2e0 40b3f9e 129e2e0 40b3f9e 129e2e0 a12a54f 40b3f9e 129e2e0 40b3f9e 129e2e0 40b3f9e 4e27cf9 40b3f9e e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 8953421 e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 4e27cf9 e35aecd 40b3f9e 966f4e1 40b3f9e 966f4e1 40b3f9e 224f37e 40b3f9e f99e269 5f6148c 40b3f9e e35aecd 40b3f9e 5f6148c 40b3f9e 35efadd 40b3f9e 6da3f79 40b3f9e 35efadd 40b3f9e 35efadd 40b3f9e 35efadd 40b3f9e |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 |
import numpy as np
import cvxpy as cp
import re
import copy
import concurrent.futures
import gradio as gr
from datetime import datetime
import random
import moviepy
from transformers import pipeline
from transformers.pipelines.audio_utils import ffmpeg_read
from moviepy.editor import (
ImageClip,
VideoFileClip,
TextClip,
CompositeVideoClip,
CompositeAudioClip,
AudioFileClip,
concatenate_videoclips,
concatenate_audioclips
)
from PIL import Image, ImageDraw, ImageFont
from moviepy.audio.AudioClip import AudioArrayClip
import subprocess
import json
import logging
import whisperx
import time
import os
import openai
from openai import OpenAI
import traceback
from TTS.api import TTS
import torch
from pydub import AudioSegment
from pyannote.audio import Pipeline
import wave
import librosa
import noisereduce as nr
import soundfile as sf
from paddleocr import PaddleOCR
import cv2
from rapidfuzz import fuzz
from tqdm import tqdm
import threading
logger = logging.getLogger(__name__)
# Configure logging
logging.basicConfig(level=logging.DEBUG, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger(__name__)
logger.info(f"MoviePy Version: {moviepy.__version__}")
# Accept license terms for Coqui XTTS
os.environ["COQUI_TOS_AGREED"] = "1"
# torch.serialization.add_safe_globals([XttsConfig])
logger.info(gr.__version__)
client = OpenAI(
api_key= os.environ.get("openAI_api_key"), # This is the default and can be omitted
)
hf_api_key = os.environ.get("hf_token")
def silence(duration, fps=44100):
"""
Returns a silent AudioClip of the specified duration.
"""
return AudioArrayClip(np.zeros((int(fps*duration), 2)), fps=fps)
def count_words_or_characters(text):
# Count non-Chinese words
non_chinese_words = len(re.findall(r'\b[a-zA-Z0-9]+\b', text))
# Count Chinese characters
chinese_chars = len(re.findall(r'[\u4e00-\u9fff]', text))
return non_chinese_words + chinese_chars
# Define the passcode
PASSCODE = "show_feedback_db"
css = """
/* Adjust row height */
.dataframe-container tr {
height: 50px !important;
}
/* Ensure text wrapping and prevent overflow */
.dataframe-container td {
white-space: normal !important;
word-break: break-word !important;
}
/* Set column widths */
[data-testid="block-container"] .scrolling-dataframe th:nth-child(1),
[data-testid="block-container"] .scrolling-dataframe td:nth-child(1) {
width: 6%; /* Start column */
}
[data-testid="block-container"] .scrolling-dataframe th:nth-child(2),
[data-testid="block-container"] .scrolling-dataframe td:nth-child(2) {
width: 47%; /* Original text */
}
[data-testid="block-container"] .scrolling-dataframe th:nth-child(3),
[data-testid="block-container"] .scrolling-dataframe td:nth-child(3) {
width: 47%; /* Translated text */
}
[data-testid="block-container"] .scrolling-dataframe th:nth-child(4),
[data-testid="block-container"] .scrolling-dataframe td:nth-child(4) {
display: none !important;
}
"""
# Function to save feedback or provide access to the database file
def handle_feedback(feedback):
feedback = feedback.strip() # Clean up leading/trailing whitespace
if not feedback:
return "Feedback cannot be empty.", None
if feedback == PASSCODE:
# Provide access to the feedback.db file
return "Access granted! Download the database file below.", "feedback.db"
else:
# Save feedback to the database
with sqlite3.connect("feedback.db") as conn:
cursor = conn.cursor()
cursor.execute("CREATE TABLE IF NOT EXISTS studio_feedback (id INTEGER PRIMARY KEY, comment TEXT)")
cursor.execute("INSERT INTO studio_feedback (comment) VALUES (?)", (feedback,))
conn.commit()
return "Thank you for your feedback!", None
def segment_background_audio(audio_path, background_audio_path="background_segments.wav", speech_audio_path="speech_segment.wav"):
"""
Uses Demucs to separate audio and extract background (non-vocal) parts.
Merges drums, bass, and other stems into a single background track.
"""
# Step 1: Run Demucs using the 4-stem model
subprocess.run([
"demucs",
"-n", "htdemucs", # 4-stem model
audio_path
], check=True)
# Step 2: Locate separated stem files
filename = os.path.splitext(os.path.basename(audio_path))[0]
stem_dir = os.path.join("separated", "htdemucs", filename)
# Step 3: Load and merge background stems
vocals = AudioSegment.from_wav(os.path.join(stem_dir, "vocals.wav"))
drums = AudioSegment.from_wav(os.path.join(stem_dir, "drums.wav"))
bass = AudioSegment.from_wav(os.path.join(stem_dir, "bass.wav"))
other = AudioSegment.from_wav(os.path.join(stem_dir, "other.wav"))
background = drums.overlay(bass).overlay(other)
# Step 4: Export the merged background
background.export(background_audio_path, format="wav")
vocals.export(speech_audio_path, format="wav")
return background_audio_path, speech_audio_path
def transcribe_video_with_speakers(video_path):
# Extract audio from video
video = VideoFileClip(video_path)
audio_path = "audio.wav"
video.audio.write_audiofile(audio_path)
logger.info(f"Audio extracted from video: {audio_path}")
segment_result, speech_audio_path = segment_background_audio(audio_path)
print(f"Saved non-speech (background) audio to local")
# Set up device
device = "cuda" if torch.cuda.is_available() else "cpu"
logger.info(f"Using device: {device}")
try:
# Load a medium model with float32 for broader compatibility
model = whisperx.load_model("large-v3", device=device, compute_type="float32")
logger.info("WhisperX model loaded")
# Transcribe
result = model.transcribe(speech_audio_path, chunk_size=4, print_progress = True)
logger.info("Audio transcription completed")
# Get the detected language
detected_language = result["language"]
logger.debug(f"Detected language: {detected_language}")
# Alignment
# model_a, metadata = whisperx.load_align_model(language_code=result["language"], device=device)
# result = whisperx.align(result["segments"], model_a, metadata, speech_audio_path, device)
# logger.info("Transcription alignment completed")
# Diarization (works independently of Whisper model size)
diarize_model = whisperx.DiarizationPipeline(use_auth_token=hf_api_key, device=device)
diarize_segments = diarize_model(speech_audio_path)
logger.info("Speaker diarization completed")
# Assign speakers
result = whisperx.assign_word_speakers(diarize_segments, result)
logger.info("Speakers assigned to transcribed segments")
except Exception as e:
logger.error(f"β WhisperX pipeline failed: {e}")
# Extract timestamps, text, and speaker IDs
transcript_with_speakers = [
{
"start": segment["start"],
"end": segment["end"],
"text": segment["text"],
"speaker": segment.get("speaker", "SPEAKER_00")
}
for segment in result["segments"]
]
# Collect audio for each speaker
speaker_audio = {}
logger.info("π Start collecting valid audio segments per speaker...")
for idx, segment in enumerate(result["segments"]):
speaker = segment.get("speaker", "SPEAKER_00")
start = segment["start"]
end = segment["end"]
if end > start and (end - start) > 0.05: # Require >50ms duration
if speaker not in speaker_audio:
speaker_audio[speaker] = [(start, end)]
else:
speaker_audio[speaker].append((start, end))
logger.debug(f"Segment {idx}: Added to speaker {speaker} [{start:.2f}s β {end:.2f}s]")
else:
logger.warning(f"β οΈ Segment {idx} discarded: invalid duration ({start:.2f}s β {end:.2f}s)")
# Collapse and truncate speaker audio
speaker_sample_paths = {}
audio_clip = AudioFileClip(speech_audio_path)
logger.info(f"π Found {len(speaker_audio)} speakers with valid segments. Start creating speaker samples...")
for speaker, segments in speaker_audio.items():
logger.info(f"πΉ Speaker {speaker}: {len(segments)} valid segments")
speaker_clips = [audio_clip.subclip(start, end) for start, end in segments]
if not speaker_clips:
logger.warning(f"β οΈ No valid audio clips for speaker {speaker}. Skipping sample creation.")
continue
if len(speaker_clips) == 1:
logger.debug(f"Speaker {speaker}: Only one clip, skipping concatenation.")
combined_clip = speaker_clips[0]
else:
logger.debug(f"Speaker {speaker}: Concatenating {len(speaker_clips)} clips.")
combined_clip = concatenate_audioclips(speaker_clips)
truncated_clip = combined_clip.subclip(0, min(30, combined_clip.duration))
logger.debug(f"Speaker {speaker}: Truncated to {truncated_clip.duration:.2f} seconds.")
# Step 4: Save the final result
sample_path = f"speaker_{speaker}_sample.wav"
truncated_clip.write_audiofile(sample_path)
speaker_sample_paths[speaker] = sample_path
logger.info(f"β
Created and saved sample for {speaker}: {sample_path}")
# Cleanup
logger.info("π§Ή Closing audio clip and removing temporary files...")
video.close()
audio_clip.close()
os.remove(speech_audio_path)
logger.info("β
Finished processing all speaker samples.")
return transcript_with_speakers, detected_language
# Function to get the appropriate translation model based on target language
# def get_translation_model(source_language, target_language):
# """
# Get the translation model based on the source and target language.
# Parameters:
# - target_language (str): The language to translate the content into (e.g., 'es', 'fr').
# - source_language (str): The language of the input content (default is 'en' for English).
# Returns:
# - str: The translation model identifier.
# """
# # List of allowable languages
# allowable_languages = ["en", "es", "fr", "zh", "de", "it", "pt", "ja", "ko", "ru", "hi", "tr"]
# # Validate source and target languages
# if source_language not in allowable_languages:
# logger.debug(f"Invalid source language '{source_language}'. Supported languages are: {', '.join(allowable_languages)}")
# # Return a default model if source language is invalid
# source_language = "en" # Default to 'en'
# if target_language not in allowable_languages:
# logger.debug(f"Invalid target language '{target_language}'. Supported languages are: {', '.join(allowable_languages)}")
# # Return a default model if target language is invalid
# target_language = "zh" # Default to 'zh'
# if source_language == target_language:
# source_language = "en" # Default to 'en'
# target_language = "zh" # Default to 'zh'
# # Return the model using string concatenation
# return f"Helsinki-NLP/opus-mt-{source_language}-{target_language}"
# def translate_single_entry(entry, translator):
# original_text = entry["text"]
# translated_text = translator(original_text)[0]['translation_text']
# return {
# "start": entry["start"],
# "original": original_text,
# "translated": translated_text,
# "end": entry["end"],
# "speaker": entry["speaker"]
# }
# def translate_text(transcription_json, source_language, target_language):
# # Load the translation model for the specified target language
# translation_model_id = get_translation_model(source_language, target_language)
# logger.debug(f"Translation model: {translation_model_id}")
# translator = pipeline("translation", model=translation_model_id)
# # Use ThreadPoolExecutor to parallelize translations
# with concurrent.futures.ThreadPoolExecutor() as executor:
# # Submit all translation tasks and collect results
# translate_func = lambda entry: translate_single_entry(entry, translator)
# translated_json = list(executor.map(translate_func, transcription_json))
# # Sort the translated_json by start time
# translated_json.sort(key=lambda x: x["start"])
# # Log the components being added to translated_json
# for entry in translated_json:
# logger.debug("Added to translated_json: start=%s, original=%s, translated=%s, end=%s, speaker=%s",
# entry["start"], entry["original"], entry["translated"], entry["end"], entry["speaker"])
# return translated_json
def update_translations(file, edited_table, source_language, target_language, process_mode):
"""
Update the translations based on user edits in the Gradio Dataframe.
"""
output_video_path = "output_video.mp4"
logger.debug(f"Editable Table: {edited_table}")
if file is None:
logger.info("No file uploaded. Please upload a video/audio file.")
return None, [], None, "No file uploaded. Please upload a video/audio file."
try:
start_time = time.time() # Start the timer
# Convert the edited_table (list of lists) back to list of dictionaries
updated_translations = [
{
"start": row["start"], # Access by column name
"original": row["original"],
"translated": row["translated"],
"end": row["end"]
}
for _, row in edited_table.iterrows()
]
translated_json = apply_adaptive_speed(updated_translations, source_language, target_language)
# Call the function to process the video with updated translations
add_transcript_voiceover(file.name, translated_json, output_video_path, process_mode)
# Calculate elapsed time
elapsed_time = time.time() - start_time
elapsed_time_display = f"Updates applied successfully in {elapsed_time:.2f} seconds."
return output_video_path, elapsed_time_display
except Exception as e:
raise ValueError(f"Error updating translations: {e}")
def create_subtitle_clip_pil(text, start_time, end_time, video_width, video_height, font_path):
try:
subtitle_width = int(video_width * 0.8)
aspect_ratio = video_height / video_width
subtitle_font_size = int(video_width // 22 if aspect_ratio > 1.2 else video_height // 24)
font = ImageFont.truetype(font_path, subtitle_font_size)
dummy_img = Image.new("RGBA", (subtitle_width, 1), (0, 0, 0, 0))
draw = ImageDraw.Draw(dummy_img)
# Word wrapping
lines = []
line = ""
for word in text.split():
test_line = f"{line} {word}".strip()
bbox = draw.textbbox((0, 0), test_line, font=font)
w = bbox[2] - bbox[0]
if w <= subtitle_width - 10:
line = test_line
else:
lines.append(line)
line = word
lines.append(line)
outline_width=2
line_heights = [draw.textbbox((0, 0), l, font=font)[3] - draw.textbbox((0, 0), l, font=font)[1] for l in lines]
total_height = sum(line_heights) + (len(lines) - 1) * 5 + 6 * outline_width
img = Image.new("RGBA", (subtitle_width, total_height), (0, 0, 0, 0))
draw = ImageDraw.Draw(img)
def draw_text_with_outline(draw, pos, text, font, fill="yellow", outline="black", outline_width = outline_width):
x, y = pos
# Draw outline
for dx in range(-outline_width, outline_width + 1):
for dy in range(-outline_width, outline_width + 1):
if dx != 0 or dy != 0:
draw.text((x + dx, y + dy), text, font=font, fill=outline)
# Draw main text
draw.text((x, y), text, font=font, fill=fill)
y = 0
for idx, line in enumerate(lines):
bbox = draw.textbbox((0, 0), line, font=font)
w = bbox[2] - bbox[0]
x = (subtitle_width - w) // 2
draw_text_with_outline(draw, (x, y), line, font)
y += line_heights[idx] + 5
img_np = np.array(img)
margin = int(video_height * 0.05)
img_clip = ImageClip(img_np) # Create the ImageClip first
image_height = img_clip.size[1]
txt_clip = (
img_clip # Use the already created clip
.set_start(start_time)
.set_duration(end_time - start_time)
.set_position(("center", video_height - image_height - margin))
.set_opacity(0.9)
)
return txt_clip
except Exception as e:
logger.error(f"β Failed to create subtitle clip: {e}")
return None
def solve_optimal_alignment(original_segments, generated_durations, total_duration):
"""
Aligns speech segments using quadratic programming. If optimization fails,
applies greedy fallback: center shorter segments, stretch longer ones.
Logs alignment results for traceability.
"""
N = len(original_segments)
d = np.array(generated_durations)
m = np.array([(seg['start'] + seg['end']) / 2 for seg in original_segments])
if N == 0 or len(generated_durations) == 0:
logger.warning("β οΈ Alignment skipped: empty segments or durations.")
return original_segments # or raise an error, depending on your app logic
try:
s = cp.Variable(N)
objective = cp.Minimize(cp.sum_squares(s + d / 2 - m))
constraints = [s[0] >= 0]
for i in range(N - 1):
constraints.append(s[i] + d[i] <= s[i + 1])
constraints.append(s[N - 1] + d[N - 1] <= total_duration)
problem = cp.Problem(objective, constraints)
problem.solve()
if s.value is None:
raise ValueError("Solver failed")
for i in range(N):
original_segments[i]['start'] = round(s.value[i], 3)
original_segments[i]['end'] = round(s.value[i] + d[i], 3)
logger.info(
f"[OPT] Segment {i}: duration={d[i]:.2f}s | start={original_segments[i]['start']:.2f}s | "
f"end={original_segments[i]['end']:.2f}s | mid={m[i]:.2f}s"
)
except Exception as e:
logger.warning(f"β οΈ Optimization failed: {e}, falling back to greedy alignment.")
for i in range(N):
orig_start = original_segments[i]['start']
orig_end = original_segments[i]['end']
orig_mid = (orig_start + orig_end) / 2
gen_duration = generated_durations[i]
orig_duration = orig_end - orig_start
if gen_duration <= orig_duration:
new_start = orig_mid - gen_duration / 2
new_end = orig_mid + gen_duration / 2
else:
extra = (gen_duration - orig_duration) / 2
new_start = orig_start - extra
new_end = orig_end + extra
if i > 0:
prev_end = original_segments[i - 1]['end']
new_start = max(new_start, prev_end + 0.01)
if i < N - 1:
next_start = original_segments[i + 1]['start']
new_end = min(new_end, next_start - 0.01)
if new_end <= new_start:
new_start = orig_start
new_end = orig_start + gen_duration
original_segments[i]['start'] = round(new_start, 3)
original_segments[i]['end'] = round(new_end, 3)
logger.info(
f"[FALLBACK] Segment {i}: duration={gen_duration:.2f}s | start={new_start:.2f}s | "
f"end={new_end:.2f}s | original_mid={orig_mid:.2f}s"
)
return original_segments
# ocr_model = None
# ocr_lock = threading.Lock()
# def init_ocr_model():
# global ocr_model
# with ocr_lock:
# if ocr_model is None:
# ocr_model = PaddleOCR(use_angle_cls=True, lang="ch")
# def find_best_subtitle_region(frame, ocr_model, region_height_ratio=0.35, num_strips=5, min_conf=0.5):
# """
# Automatically identifies the best subtitle region in a video frame using OCR confidence.
# Parameters:
# - frame: full video frame (BGR np.ndarray)
# - ocr_model: a loaded PaddleOCR model
# - region_height_ratio: portion of image height to scan (from bottom up)
# - num_strips: how many horizontal strips to evaluate
# - min_conf: minimum average confidence to consider a region valid
# Returns:
# - crop_region: the cropped image region with highest OCR confidence
# - region_box: (y_start, y_end) of the region in the original frame
# """
# height, width, _ = frame.shape
# region_height = int(height * region_height_ratio)
# base_y_start = height - region_height
# strip_height = region_height // num_strips
# best_score = -1
# best_crop = None
# best_bounds = (0, height)
# for i in range(num_strips):
# y_start = base_y_start + i * strip_height
# y_end = y_start + strip_height
# strip = frame[y_start:y_end, :]
# try:
# result = ocr_model.ocr(strip, cls=True)
# if not result or not result[0]:
# continue
# total_score = sum(line[1][1] for line in result[0])
# avg_score = total_score / len(result[0])
# if avg_score > best_score:
# best_score = avg_score
# best_crop = strip
# best_bounds = (y_start, y_end)
# except Exception as e:
# continue # Fail silently on OCR issues
# if best_score >= min_conf and best_crop is not None:
# return best_crop, best_bounds
# else:
# # Fallback to center-bottom strip
# fallback_y = height - int(height * 0.2)
# return frame[fallback_y:, :], (fallback_y, height)
# def ocr_frame_worker(args, min_confidence=0.7):
# frame_idx, frame_time, frame = args
# init_ocr_model() # Load model in thread-safe way
# if frame is None or frame.size == 0 or not isinstance(frame, np.ndarray):
# return {"time": frame_time, "text": ""}
# if frame.dtype != np.uint8:
# frame = frame.astype(np.uint8)
# try:
# result = ocr_model.ocr(frame, cls=True)
# lines = result[0] if result else []
# texts = [line[1][0] for line in lines if line[1][1] >= min_confidence]
# combined_text = " ".join(texts).strip()
# return {"time": frame_time, "text": combined_text}
# except Exception as e:
# print(f"β οΈ OCR failed at {frame_time:.2f}s: {e}")
# return {"time": frame_time, "text": ""}
# def frame_is_in_audio_segments(frame_time, audio_segments, tolerance=0.2):
# for segment in audio_segments:
# start, end = segment["start"], segment["end"]
# if (start - tolerance) <= frame_time <= (end + tolerance):
# return True
# return False
# def extract_ocr_subtitles_parallel(video_path, transcription_json, interval_sec=0.5, num_workers=4):
# cap = cv2.VideoCapture(video_path)
# fps = cap.get(cv2.CAP_PROP_FPS)
# frames = []
# frame_idx = 0
# success, frame = cap.read()
# while success:
# if frame_idx % int(fps * interval_sec) == 0:
# frame_time = frame_idx / fps
# if frame_is_in_audio_segments(frame_time, transcription_json):
# frames.append((frame_idx, frame_time, frame.copy()))
# success, frame = cap.read()
# frame_idx += 1
# cap.release()
# ocr_results = []
# ocr_failures = 0 # Count OCR failures
# with concurrent.futures.ThreadPoolExecutor(max_workers=num_workers) as executor:
# futures = [executor.submit(ocr_frame_worker, frame) for frame in frames]
# for f in tqdm(concurrent.futures.as_completed(futures), total=len(futures)):
# try:
# result = f.result()
# if result["text"]:
# ocr_results.append(result)
# except Exception as e:
# ocr_failures += 1
# logger.info(f"β
OCR extraction completed: {len(ocr_results)} frames successful, {ocr_failures} frames failed.")
# return ocr_results
# def collapse_ocr_subtitles(ocr_json, text_similarity_threshold=90):
# collapsed = []
# current = None
# for entry in ocr_json:
# time = entry["time"]
# text = entry["text"]
# if not current:
# current = {"start": time, "end": time, "text": text}
# continue
# sim = fuzz.ratio(current["text"], text)
# if sim >= text_similarity_threshold:
# current["end"] = time
# logger.debug(f"MERGED: Current end extended to {time:.2f}s for text: '{current['text'][:50]}...' (Similarity: {sim})")
# else:
# logger.debug(f"NOT MERGING (Similarity: {sim} < Threshold: {text_similarity_threshold}):")
# logger.debug(f" Previous segment: {current['start']:.2f}s - {current['end']:.2f}s: '{current['text'][:50]}...'")
# logger.debug(f" New segment: {time:.2f}s: '{text[:50]}...'")
# collapsed.append(current)
# current = {"start": time, "end": time, "text": text}
# if current:
# collapsed.append(current)
# logger.info(f"β
OCR subtitles collapsed into {len(collapsed)} segments.")
# for idx, seg in enumerate(collapsed):
# logger.debug(f"[OCR Collapsed {idx}] {seg['start']:.2f}s - {seg['end']:.2f}s: {seg['text'][:50]}...")
# return collapsed
# def merge_speaker_and_time_from_whisperx(
# ocr_json,
# whisperx_json,
# replace_threshold=90,
# time_tolerance=1.0
# ):
# merged = []
# used_whisperx = set()
# whisperx_used_flags = [False] * len(whisperx_json)
# # Step 1: Attempt to match each OCR entry to a WhisperX entry
# for ocr in ocr_json:
# ocr_start, ocr_end = ocr["start"], ocr["end"]
# ocr_text = ocr["text"]
# best_match = None
# best_score = -1
# best_idx = None
# for idx, wx in enumerate(whisperx_json):
# wx_start, wx_end = wx["start"], wx["end"]
# wx_text = wx["text"]
# # Check for time overlap
# overlap = not (ocr_end < wx_start - time_tolerance or ocr_start > wx_end + time_tolerance)
# if not overlap:
# continue
# sim = fuzz.ratio(ocr_text, wx_text)
# if sim > best_score:
# best_score = sim
# best_match = wx
# best_idx = idx
# if best_match and best_score >= replace_threshold:
# # Replace WhisperX segment with higher quality OCR text
# new_segment = copy.deepcopy(best_match)
# new_segment["text"] = ocr_text
# new_segment["ocr_replaced"] = True
# new_segment["ocr_similarity"] = best_score
# whisperx_used_flags[best_idx] = True
# merged.append(new_segment)
# else:
# # No replacement, check if this OCR is outside WhisperX time coverage
# covered = any(
# abs((ocr_start + ocr_end)/2 - (wx["start"] + wx["end"])/2) < time_tolerance
# for wx in whisperx_json
# )
# if not covered:
# new_segment = copy.deepcopy(ocr)
# new_segment["ocr_added"] = True
# new_segment["speaker"] = "UNKNOWN"
# merged.append(new_segment)
# # Step 2: Add untouched WhisperX segments
# for idx, wx in enumerate(whisperx_json):
# if not whisperx_used_flags[idx]:
# merged.append(wx)
# # Step 3: Sort all merged segments
# merged = sorted(merged, key=lambda x: x["start"])
# return merged
# --- Function Definitions ---
def process_segment_with_gpt(segment, source_lang, target_lang, model="gpt-4", openai_client=None):
"""
Processes a single text segment: restores punctuation and translates using an OpenAI GPT model.
"""
if openai_client is None:
segment_identifier = f"{segment.get('start', 'N/A')}-{segment.get('end', 'N/A')}"
logger.error(f"β OpenAI client was not provided for segment {segment_identifier}. Cannot process.")
return {
"start": segment.get("start"),
"end": segment.get("end"),
"speaker": segment.get("speaker", "SPEAKER_00"),
"original": segment["text"],
"translated": "[ERROR: OpenAI client not provided]"
}
original_text = segment["text"]
segment_id = f"{segment['start']}-{segment['end']}" # Create a unique ID for this segment for easier log tracking
logger.debug(
f"Starting processing for segment {segment_id}. "
f"Original text preview: '{original_text[:100]}{'...' if len(original_text) > 100 else ''}'"
)
prompt = (
f"You are a multilingual assistant. Given the following text in {source_lang}, "
f"1) restore punctuation, and 2) translate it into {target_lang}.\n\n"
f"Text:\n{original_text}\n\n"
f"Return in JSON format:\n"
f'{{"punctuated": "...", "translated": "..."}}'
)
try:
logger.debug(f"Sending request to OpenAI model '{model}' for segment {segment_id}...")
response = openai_client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": prompt}],
temperature=0.3
)
content = response.choices[0].message.content.strip()
# --- NEW LOGIC: Clean markdown code block fences from the response ---
cleaned_content = content
if content.startswith("```") and content.endswith("```"):
# Attempt to find the actual JSON object within the markdown fence
json_start_index = content.find('{')
json_end_index = content.rfind('}')
if json_start_index != -1 and json_end_index != -1 and json_end_index > json_start_index:
cleaned_content = content[json_start_index : json_end_index + 1]
logger.debug(f"Removed markdown fences for segment {segment_id}. Extracted JSON portion.")
else:
logger.warning(
f"β οΈ Content starts/ends with '```' but a valid JSON object ({{...}}) was not found within "
f"fences for segment {segment_id}. Attempting to parse raw content. Raw content: '{content}'"
)
# --- END NEW LOGIC ---
logger.debug(
f"Attempting to parse JSON for segment {segment_id}. "
f"Content for parsing preview: '{cleaned_content[:200]}{'...' if len(cleaned_content) > 200 else ''}'"
)
result_json = {}
try:
result_json = json.loads(cleaned_content)
except json.JSONDecodeError as e:
logger.warning(
f"β οΈ Failed to parse JSON response for segment {segment_id}. Error: {e}. "
f"Content attempted to parse: '{cleaned_content}'" # Log cleaned content here
)
punctuated_text = original_text
translated_text = "" # Return empty translated text on parsing failure
else:
punctuated_text = result_json.get("punctuated", original_text)
translated_text = result_json.get("translated", "")
logger.info(
f"β
Successfully processed segment {segment_id}. "
f"Punctuated preview: '{punctuated_text[:50]}{'...' if len(punctuated_text) > 50 else ''}', "
f"Translated preview: '{translated_text[:50]}{'...' if len(translated_text) > 50 else ''}'"
)
return {
"start": segment["start"],
"end": segment["end"],
"speaker": segment.get("speaker", "SPEAKER_00"),
"original": punctuated_text,
"translated": translated_text
}
except Exception as e:
logger.error(
f"β An unexpected error occurred for segment {segment_id}: {e}",
exc_info=True # This logs the full traceback
)
return {
"start": segment["start"],
"end": segment["end"],
"speaker": segment.get("speaker", "SPEAKER_00"),
"original": original_text,
"translated": "[ERROR: Processing failed]"
}
def punctuate_and_translate_parallel(transcription_json, source_lang="zh", target_lang="en", model="gpt-4o", max_workers=5, openai_client=None):
"""
Orchestrates parallel punctuation restoration and translation of multiple segments
using a ThreadPoolExecutor.
"""
if not transcription_json:
logger.warning("No segments provided in transcription_json for parallel processing. Returning an empty list.")
return []
logger.info(f"Starting parallel punctuation and translation for {len(transcription_json)} segments.")
logger.info(
f"Configuration: Model='{model}', Source Language='{source_lang}', "
f"Target Language='{target_lang}', Max Workers={max_workers}."
)
results = []
with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as executor:
# Submit each segment for processing, ensuring the openai_client is passed to each worker
futures = {
executor.submit(process_segment_with_gpt, seg, source_lang, target_lang, model, openai_client): seg
for seg in transcription_json
}
logger.info(f"All {len(futures)} segments have been submitted to the thread pool.")
# Asynchronously collect results as they complete
for i, future in enumerate(concurrent.futures.as_completed(futures)):
segment = futures[future] # Retrieve the original segment data for logging context
segment_id = f"{segment['start']}-{segment['end']}"
try:
result = future.result() # This will re-raise any exception from the worker thread
results.append(result)
logger.debug(f"Collected result for segment {segment_id}. ({i + 1}/{len(futures)} completed)")
except Exception as exc:
# This catch block is for rare cases where the future itself fails to yield a result,
# or an exception was not caught within `process_segment_with_gpt`.
logger.error(
f"Unhandled exception encountered while retrieving result for segment {segment_id}: {exc}",
exc_info=True
)
# Ensure a placeholder result is added even if future retrieval fails
results.append({
"start": segment.get("start"),
"end": segment.get("end"),
"speaker": segment.get("speaker", "SPEAKER_00"),
"original": segment["text"],
"translated": "[ERROR: Unhandled exception in parallel processing]"
})
logger.info("π Parallel processing complete. All results collected.")
return results
# def merge_speaker_and_time_from_whisperx(ocr_json, whisperx_json, text_sim_threshold=80, replace_threshold=90):
# merged = []
# used_whisperx = set()
# for ocr in ocr_json:
# ocr_start = ocr["start"]
# ocr_end = ocr["end"]
# ocr_text = ocr["text"]
# best_match = None
# best_score = -1
# best_idx = None
# for idx, wx in enumerate(whisperx_json):
# wx_start, wx_end = wx["start"], wx["end"]
# wx_text = wx["text"]
# if idx in used_whisperx:
# continue # Already matched
# time_center_diff = abs((ocr_start + ocr_end)/2 - (wx_start + wx_end)/2)
# if time_center_diff > 3:
# continue
# sim = fuzz.ratio(ocr_text, wx_text)
# if sim > best_score:
# best_score = sim
# best_match = wx
# best_idx = idx
# new_entry = copy.deepcopy(ocr)
# if best_match:
# new_entry["speaker"] = best_match.get("speaker", "UNKNOWN")
# new_entry["ocr_similarity"] = best_score
# if best_score >= replace_threshold:
# new_entry["start"] = best_match["start"]
# new_entry["end"] = best_match["end"]
# used_whisperx.add(best_idx) # Mark used
# else:
# new_entry["speaker"] = "UNKNOWN"
# new_entry["ocr_similarity"] = None
# merged.append(new_entry)
# return merged
def realign_ocr_segments(merged_ocr_json, min_gap=0.2):
"""
Realign OCR segments to avoid overlaps using midpoint-based adjustment.
"""
merged_ocr_json = sorted(merged_ocr_json, key=lambda x: x["start"])
for i in range(1, len(merged_ocr_json)):
prev = merged_ocr_json[i - 1]
curr = merged_ocr_json[i]
# If current overlaps with previous, adjust
if curr["start"] < prev["end"] + min_gap:
midpoint = (prev["end"] + curr["start"]) / 2
prev["end"] = round(midpoint - min_gap / 2, 3)
curr["start"] = round(midpoint + min_gap / 2, 3)
# Prevent negative durations
if curr["start"] >= curr["end"]:
curr["end"] = round(curr["start"] + 0.3, 3)
return merged_ocr_json
def post_edit_transcribed_segments(transcription_json, video_path,
interval_sec=0.5,
text_similarity_threshold=80,
time_tolerance=1.0,
num_workers=4):
"""
Given WhisperX transcription (transcription_json) and video,
use OCR subtitles to post-correct and safely insert missing captions.
"""
# Step 1: Extract OCR subtitles (only near audio segments)
ocr_json = extract_ocr_subtitles_parallel(
video_path,
transcription_json,
interval_sec=interval_sec,
num_workers=num_workers
)
# Step 2: Collapse repetitive OCR
collapsed_ocr = collapse_ocr_subtitles(ocr_json, text_similarity_threshold=90)
# Step 3: Merge and realign OCR segments.
ocr_merged = merge_speaker_and_time_from_whisperx(collapsed_ocr, transcription_json)
ocr_realigned = realign_ocr_segments(ocr_merged)
logger.info(f"β
Final merged and realigned OCR: {len(ocr_realigned)} segments")
return ocr_realigned
def process_entry(entry, i, tts_model, video_width, video_height, process_mode, target_language, font_path, speaker_sample_paths=None):
logger.debug(f"Processing entry {i}: {entry}")
error_message = None
try:
txt_clip = create_subtitle_clip_pil(entry["translated"], entry["start"], entry["end"], video_width, video_height, font_path)
except Exception as e:
error_message = f"β Failed to create subtitle clip for entry {i}: {e}"
logger.error(error_message)
txt_clip = None
audio_segment = None
actual_duration = 0.0
if process_mode > 1:
try:
segment_audio_path = f"segment_{i}_voiceover.wav"
desired_duration = entry["end"] - entry["start"]
desired_speed = entry['speed'] #calibrated_speed(entry['translated'], desired_duration)
speaker = entry.get("speaker", "SPEAKER_00")
speaker_wav_path = f"speaker_{speaker}_sample.wav"
if process_mode > 2 and speaker_wav_path and os.path.exists(speaker_wav_path) and target_language in tts_model.synthesizer.tts_model.language_manager.name_to_id.keys():
generate_voiceover_clone(entry['translated'], tts_model, desired_speed, target_language, speaker_wav_path, segment_audio_path)
else:
generate_voiceover_OpenAI(entry['translated'], target_language, desired_speed, segment_audio_path)
if not segment_audio_path or not os.path.exists(segment_audio_path):
raise FileNotFoundError(f"Voiceover file not generated at: {segment_audio_path}")
audio_clip = AudioFileClip(segment_audio_path)
actual_duration = audio_clip.duration
audio_segment = audio_clip # Do not set start here, alignment happens later
except Exception as e:
err = f"β Failed to generate audio segment for entry {i}: {e}"
logger.error(err)
error_message = error_message + " | " + err if error_message else err
audio_segment = None
return i, txt_clip, audio_segment, actual_duration, error_message
def add_transcript_voiceover(video_path, translated_json, output_path, process_mode, target_language="en", speaker_sample_paths=None, background_audio_path="background_segments.wav"):
video = VideoFileClip(video_path)
font_path = "./NotoSansSC-Regular.ttf"
text_clips = []
audio_segments = []
actual_durations = []
error_messages = []
if process_mode > 2:
global tts_model
if tts_model is None:
try:
print("π Loading XTTS model...")
from TTS.api import TTS
tts_model = TTS(model_name="tts_models/multilingual/multi-dataset/your_tts")
print("β
XTTS model loaded successfully.")
except Exception as e:
print("β Error loading XTTS model:")
traceback.print_exc()
return f"Error loading XTTS model: {e}"
with concurrent.futures.ThreadPoolExecutor() as executor:
# futures = [executor.submit(process_entry, entry, i, tts_model, video.w, video.h, process_mode, target_language, font_path, speaker_sample_paths)
# # for i, entry in enumerate(translated_json)]
# results = []
# for future in concurrent.futures.as_completed(futures):
# try:
# i, txt_clip, audio_segment, actual_duration, error = future.result()
# results.append((i, txt_clip, audio_segment, actual_duration))
# if error:
# error_messages.append(f"[Entry {i}] {error}")
# except Exception as e:
# err = f"β Unexpected error in future result: {e}"
# error_messages.append(err)
# Use dict as a placeholder, any failure will leave a (None, None, 0)
futures = {
executor.submit(
process_entry, entry, idx, tts_model, video.w, video.h,
process_mode, target_language, font_path, speaker_sample_paths
): idx
for idx, entry in enumerate(translated_json)
}
# Give each entry a placeholder first to prevent overstepping boundaries
result_map = {idx: (None, None, 0) for idx in range(len(translated_json))}
for future in concurrent.futures.as_completed(futures):
idx = futures[future]
try:
_idx, txt, aud, dur, err = future.result()
result_map[idx] = (txt, aud, dur)
if err:
error_messages.append(f"[Entry {idx}] {err}")
except Exception as e:
# Threads that throw errors also need to take up space to prevent the list index from going out of range
error_messages.append(f"[Entry {idx}] unexpected error: {e}")
# results.sort(key=lambda x: x[0])
# text_clips = [clip for _, clip, _, _ in results if clip]
# generated_durations = [dur for _, _, _, dur in results if dur > 0]
# Sort and filter together
results.sort(key=lambda x: x[0])
filtered = [(translated_json[i], txt, aud, dur) for i, txt, aud, dur in results if dur > 0]
translated_json = [entry for entry, _, _, _ in filtered]
generated_durations = [dur for _, _, _, dur in filtered]
# Align using optimization (modifies translated_json in-place)
if generated_durations:
translated_json = solve_optimal_alignment(translated_json, generated_durations, video.duration)
else:
logger.warning("No generated audio; skip alignment optimisation.")
# Set aligned timings
# audio_segments = []
# for i, entry in enumerate(translated_json):
# segment = results[i][2] # AudioFileClip
# if segment:
# segment = segment.set_start(entry['start']).set_duration(entry['end'] - entry['start'])
# audio_segments.append(segment)
audio_segments = []
for i, entry in enumerate(translated_json):
_, seg, _dur = result_map[i] # seg is AudioFileClip
if seg:
audio_segments.append(
seg.set_start(entry["start"]).set_duration(entry["end"] - entry["start"])
)
final_video = CompositeVideoClip([video] + text_clips)
if process_mode > 1 and audio_segments:
try:
voice_audio = CompositeAudioClip(audio_segments).set_duration(video.duration)
if background_audio_path and os.path.exists(background_audio_path):
background_audio = AudioFileClip(background_audio_path).set_duration(video.duration)
final_audio = CompositeAudioClip([voice_audio, background_audio])
else:
final_audio = voice_audio
final_video = final_video.set_audio(final_audio)
except Exception as e:
print(f"β Failed to set audio: {e}")
final_video.write_videofile(output_path, codec="libx264", audio_codec="aac")
return error_messages
def generate_voiceover_OpenAI(full_text, language, desired_speed, output_audio_path):
"""
Generate voiceover from translated text for a given language using OpenAI TTS API.
"""
# Define the voice based on the language (for now, use 'alloy' as default)
voice = "alloy" # Adjust based on language if needed
# Define the model (use tts-1 for real-time applications)
model = "tts-1"
max_retries = 3
retry_count = 0
while retry_count < max_retries:
try:
# Create the speech using OpenAI TTS API
response = client.audio.speech.create(
model=model,
voice=voice,
input=full_text,
speed=desired_speed
)
# Save the audio to the specified path
with open(output_audio_path, 'wb') as f:
for chunk in response.iter_bytes():
f.write(chunk)
logging.info(f"Voiceover generated successfully for {output_audio_path}")
break
except Exception as e:
retry_count += 1
logging.error(f"Error generating voiceover (retry {retry_count}/{max_retries}): {e}")
time.sleep(5) # Wait 5 seconds before retrying
if retry_count == max_retries:
raise ValueError(f"Failed to generate voiceover after {max_retries} retries.")
def generate_voiceover_clone(full_text, tts_model, desired_speed, target_language, speaker_wav_path, output_audio_path):
try:
tts_model.tts_to_file(
text=full_text,
speaker_wav=speaker_wav_path,
language=target_language,
file_path=output_audio_path,
speed=desired_speed,
split_sentences=True
)
msg = (
f"β
Voice cloning completed successfully. "
f"[Speaker Wav: {speaker_wav_path}] [Speed: {desired_speed}]"
)
logger.info(msg)
return output_audio_path, msg, None
except Exception as e:
generate_voiceover_OpenAI(full_text, target_language, desired_speed, output_audio_path)
err_msg = f"β An error occurred: {str(e)}, fallback to premium voice"
logger.error(traceback.format_exc())
return None, err_msg, err_msg
def apply_adaptive_speed(translated_json_raw, source_language, target_language, k=3.0, default_prior_speed=5.0):
"""
Adds `speed` (relative, 1.0 = normal speed) and `target_duration` (sec) to each segment
using shrinkage-based estimation, language stretch ratios, and optional style modifiers.
Speeds are clamped to [0.85, 1.7] to avoid unnatural TTS behavior.
"""
translated_json = copy.deepcopy(translated_json_raw)
# Prior average speech speeds by (category, target language)
priors = {
("drama", "en"): 5.0,
("drama", "zh"): 4.5,
("tutorial", "en"): 5.2,
("tutorial", "zh"): 4.8,
("shortplay", "en"): 5.1,
("shortplay", "zh"): 4.7,
}
# Adjustment ratio based on language pair (source β target)
lang_ratio = {
("zh", "en"): 0.85,
("en", "zh"): 1.15,
("zh", "jp"): 1.05,
("en", "ja"): 0.9,
}
# Optional style modulation factor
style_modifiers = {
"dramatic": 0.9,
"urgent": 1.1,
"neutral": 1.0
}
for idx, entry in enumerate(translated_json):
start, end = float(entry.get("start", 0)), float(entry.get("end", 0))
duration = max(0.1, end - start)
original_text = entry.get("original", "")
translated_text = entry.get("translated", "")
category = entry.get("category", "drama")
source_lang = source_language
target_lang = target_language
style = entry.get("style", "neutral").lower()
# Observed speed from original
base_text = original_text or translated_text
obs_speed = len(base_text) / duration
# Prior speed
prior_speed = priors.get((category, target_lang), default_prior_speed)
# Shrinkage
shrink_speed = (duration * obs_speed + k * prior_speed) / (duration + k)
# Language pacing adjustment
ratio = lang_ratio.get((source_lang, target_lang), 1.0)
adjusted_speed = shrink_speed * ratio
# Style modulation
mod = style_modifiers.get(style, 1.0)
adjusted_speed *= mod
# Final relative speed (normalized to prior)
relative_speed = adjusted_speed / prior_speed
# Clamp relative speed to [0.85, 1.7]
relative_speed = max(0.85, min(1.7, relative_speed))
# Compute target duration for synthesis
target_chars = len(translated_text)
target_duration = round(target_chars / adjusted_speed, 2)
# Logging
logger.info(
f"[Segment {idx}] dur={duration:.2f}s | obs_speed={obs_speed:.2f} | prior={prior_speed:.2f} | "
f"shrinked={shrink_speed:.2f} | lang_ratio={ratio} | style_mod={mod} | "
f"adj_speed={adjusted_speed:.2f} | rel_speed={relative_speed:.2f} | "
f"target_dur={target_duration:.2f}s"
)
entry["speed"] = round(relative_speed, 3)
entry["target_duration"] = target_duration
return translated_json
def calibrated_speed(text, desired_duration):
"""
Compute a speed factor to help TTS fit audio into desired duration,
using a simple truncated linear function of characters per second.
"""
char_count = len(text.strip())
if char_count == 0 or desired_duration <= 0:
return 1.0 # fallback
cps = char_count / desired_duration # characters per second
# Truncated linear mapping
if cps < 14:
return 1.0
elif cps > 25.2:
return 1.7
else:
slope = (1.7 - 1.0) / (25.2 - 14)
return 1.0 + slope * (cps - 14)
def upload_and_manage(file, target_language, process_mode):
if file is None:
logger.info("No file uploaded. Please upload a video/audio file.")
return None, [], None, "No file uploaded. Please upload a video/audio file."
try:
start_time = time.time() # Start the timer
logger.info(f"Started processing file: {file.name}")
# Define paths for audio and output files
audio_path = "audio.wav"
output_video_path = "output_video.mp4"
voiceover_path = "voiceover.wav"
logger.info(f"Using audio path: {audio_path}, output video path: {output_video_path}, voiceover path: {voiceover_path}")
# Step 1: Transcribe audio from uploaded media file and get timestamps
logger.info("Transcribing audio...")
transcription_json, source_language = transcribe_video_with_speakers(file.name)
logger.info(f"Transcription completed. Detected source language: {source_language}")
translated_json_raw = punctuate_and_translate_parallel(transcription_json, source_language, target_language, openai_client = client)
# Step 2: Translate the transcription
# logger.info(f"Translating transcription from {source_language} to {target_language}...")
# translated_json_raw = translate_text(transcription_json_merged, )
logger.info(f"Translation completed. Number of translated segments: {len(translated_json_raw)}")
translated_json = apply_adaptive_speed(translated_json_raw, source_language, target_language)
# Step 3: Add transcript to video based on timestamps
logger.info("Adding translated transcript to video...")
add_transcript_voiceover(file.name, translated_json, output_video_path, process_mode, target_language)
logger.info(f"Transcript added to video. Output video saved at {output_video_path}")
# Convert translated JSON into a format for the editable table
logger.info("Converting translated JSON into editable table format...")
editable_table = [
[float(entry["start"]), entry["original"], entry["translated"], float(entry["end"]), entry["speaker"]]
for entry in translated_json
]
# Calculate elapsed time
elapsed_time = time.time() - start_time
elapsed_time_display = f"Processing completed in {elapsed_time:.2f} seconds."
logger.info(f"Processing completed in {elapsed_time:.2f} seconds.")
return source_language, editable_table, output_video_path, elapsed_time_display
except Exception as e:
logger.error(f"An error occurred: {str(e)}")
return None, [], None, f"An error occurred: {str(e)}"
# Gradio Interface with Tabs
def build_interface():
with gr.Blocks(css=css) as demo:
gr.Markdown("## Video Localization")
with gr.Row():
with gr.Column(scale=4):
file_input = gr.File(label="Upload Video/Audio File")
language_input = gr.Dropdown(["en", "es", "fr", "zh"], label="Select Language")
source_language_display = gr.Textbox(label="Detected Source Language", interactive=False)
process_mode = gr.Radio(
choices=[("Transcription Only", 1), ("Transcription with Premium Voice", 2), ("Transcription with Voice Clone", 3)],
label="Choose Processing Type",
value=1
)
submit_button = gr.Button("Post and Process")
with gr.Column(scale=8):
gr.Markdown("## Edit Translations")
# Editable JSON Data
editable_table = gr.Dataframe(
value=[], # Default to an empty list to avoid undefined values
headers=["start", "original", "translated", "end", "speaker"],
datatype=["number", "str", "str", "number", "str"],
row_count=1, # Initially empty
col_count=5,
interactive=[False, True, True, False, False], # Control editability
label="Edit Translations",
wrap=True # Enables text wrapping if supported
)
save_changes_button = gr.Button("Save Changes")
processed_video_output = gr.File(label="Download Processed Video", interactive=True) # Download button
elapsed_time_display = gr.Textbox(label="Elapsed Time", lines=1, interactive=False)
with gr.Column(scale=1):
gr.Markdown("**Feedback**")
feedback_input = gr.Textbox(
placeholder="Leave your feedback here...",
label=None,
lines=3,
)
feedback_btn = gr.Button("Submit Feedback")
response_message = gr.Textbox(label=None, lines=1, interactive=False)
db_download = gr.File(label="Download Database File", visible=False)
# Link the feedback handling
def feedback_submission(feedback):
message, file_path = handle_feedback(feedback)
if file_path:
return message, gr.update(value=file_path, visible=True)
return message, gr.update(visible=False)
save_changes_button.click(
update_translations,
inputs=[file_input, editable_table, source_language_display, language_input, process_mode],
outputs=[processed_video_output, elapsed_time_display]
)
submit_button.click(
upload_and_manage,
inputs=[file_input, language_input, process_mode],
outputs=[source_language_display, editable_table, processed_video_output, elapsed_time_display]
)
# Connect submit button to save_feedback_db function
feedback_btn.click(
feedback_submission,
inputs=[feedback_input],
outputs=[response_message, db_download]
)
return demo
tts_model = None
# Launch the Gradio interface
demo = build_interface()
demo.launch() |