File size: 48,280 Bytes
b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 d689086 b5df735 |
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 |
"""
Distributed Transcription Service
Handles audio transcription with true distributed processing across multiple Modal containers
Enhanced with intelligent audio segmentation capabilities
"""
import asyncio
import aiohttp
import base64
import os
import tempfile
import subprocess
import json
from pathlib import Path
from typing import Dict, Any, List, Tuple
from concurrent.futures import ThreadPoolExecutor
import time
import re
import ffmpeg
import torch
from .transcription_service import TranscriptionService
class DistributedTranscriptionService:
"""Service for handling distributed audio transcription across multiple Modal containers"""
def __init__(self, cache_dir: str = "/tmp"):
self.cache_dir = cache_dir
self.transcription_service = TranscriptionService(cache_dir)
def split_audio_by_time(self, audio_file_path: str, chunk_duration: int = 60) -> List[Dict[str, Any]]:
"""Split audio into time-based chunks"""
try:
# Get audio duration using ffprobe
duration_cmd = [
"ffprobe", "-v", "quiet", "-show_entries", "format=duration",
"-of", "csv=p=0", audio_file_path
]
result = subprocess.run(duration_cmd, capture_output=True, text=True, check=True)
total_duration = float(result.stdout.strip())
chunks = []
start_time = 0.0
chunk_index = 0
while start_time < total_duration:
end_time = min(start_time + chunk_duration, total_duration)
actual_duration = end_time - start_time
# Skip very short chunks (less than 5 seconds)
if actual_duration < 5.0:
break
chunk_filename = f"chunk_{chunk_index:03d}.wav"
chunks.append({
"chunk_index": chunk_index,
"start_time": start_time,
"end_time": end_time,
"duration": actual_duration,
"filename": chunk_filename
})
start_time = end_time
chunk_index += 1
print(f"π Split audio into {len(chunks)} time-based chunks")
return chunks
except Exception as e:
print(f"β Error splitting audio by time: {e}")
return []
def split_audio_by_silence(
self,
audio_file_path: str,
min_segment_length: float = 30.0,
min_silence_length: float = 1.0,
max_segment_length: float = 120.0
) -> List[Dict[str, Any]]:
"""
Intelligently split audio using FFmpeg's silencedetect filter
Enhanced from AudioProcessingService
"""
try:
silence_end_re = re.compile(
r" silence_end: (?P<end>[0-9]+(\.?[0-9]*)) \| silence_duration: (?P<dur>[0-9]+(\.?[0-9]*))"
)
# Get audio duration
metadata = ffmpeg.probe(audio_file_path)
total_duration = float(metadata["format"]["duration"])
print(f"π΅ Audio duration: {total_duration:.2f}s")
print(f"π Detecting silence with min_silence_length={min_silence_length}s...")
# Use silence detection filter
cmd = [
"ffmpeg", "-i", audio_file_path,
"-af", f"silencedetect=noise=-30dB:duration={min_silence_length}",
"-f", "null", "-"
]
process = subprocess.Popen(
cmd,
stderr=subprocess.PIPE,
stdout=subprocess.PIPE,
text=True
)
segments = []
cur_start = 0.0
chunk_index = 0
# Process silence detection output
for line in process.stderr:
match = silence_end_re.search(line)
if match:
silence_end = float(match.group("end"))
silence_dur = float(match.group("dur"))
split_at = silence_end - (silence_dur / 2)
segment_duration = split_at - cur_start
# Skip segments that are too short
if segment_duration < min_segment_length:
continue
# Split long segments
if segment_duration > max_segment_length:
# Split into multiple smaller segments
sub_start = cur_start
while sub_start < split_at:
sub_end = min(sub_start + max_segment_length, split_at)
sub_duration = sub_end - sub_start
if sub_duration >= min_segment_length:
segments.append({
"chunk_index": chunk_index,
"start_time": sub_start,
"end_time": sub_end,
"duration": sub_duration,
"filename": f"silence_chunk_{chunk_index:03d}.wav",
"segmentation_type": "silence_based"
})
chunk_index += 1
sub_start = sub_end
else:
segments.append({
"chunk_index": chunk_index,
"start_time": cur_start,
"end_time": split_at,
"duration": segment_duration,
"filename": f"silence_chunk_{chunk_index:03d}.wav",
"segmentation_type": "silence_based"
})
chunk_index += 1
cur_start = split_at
process.wait()
# Handle the last segment
if total_duration > cur_start:
remaining_duration = total_duration - cur_start
if remaining_duration >= min_segment_length:
segments.append({
"chunk_index": chunk_index,
"start_time": cur_start,
"end_time": total_duration,
"duration": remaining_duration,
"filename": f"silence_chunk_{chunk_index:03d}.wav",
"segmentation_type": "silence_based"
})
print(f"π― Silence-based segmentation created {len(segments)} segments")
return segments
except Exception as e:
print(f"β οΈ Silence-based segmentation failed: {e}")
# Fallback to time-based segmentation
print("π Falling back to time-based segmentation...")
return self.split_audio_by_time(audio_file_path, chunk_duration=60)
def choose_segmentation_strategy(
self,
audio_file_path: str,
use_intelligent_segmentation: bool = True,
chunk_duration: int = 60
) -> List[Dict[str, Any]]:
"""
Choose the best segmentation strategy based on audio characteristics
"""
try:
# Get audio metadata
metadata = ffmpeg.probe(audio_file_path)
duration = float(metadata["format"]["duration"])
print(f"ποΈ Choosing segmentation strategy for {duration:.2f}s audio...")
# For short audio (< 30s), use single processing
if duration < 30:
print("π Audio is short, using single chunk")
return [{
"chunk_index": 0,
"start_time": 0.0,
"end_time": duration,
"duration": duration,
"filename": "single_chunk.wav",
"segmentation_type": "single"
}]
# For longer audio, choose based on user preference
if use_intelligent_segmentation:
print("π§ Using intelligent silence-based segmentation")
segments = self.split_audio_by_silence(
audio_file_path,
min_segment_length=30.0,
min_silence_length=1.0,
max_segment_length=120.0
)
# NEW: Check if silence-based segmentation failed for long audio
if duration > 180 and len(segments) == 1: # Audio > 3 minutes with only 1 segment
print(f"β οΈ Silence-based segmentation created only 1 segment for {duration:.2f}s audio")
print("π Falling back to 3-minute time-based segmentation for better processing efficiency")
return self.split_audio_by_time(audio_file_path, chunk_duration=180) # 3-minute chunks
# If silence-based segmentation didn't work well, fallback to time-based
if len(segments) == 0 or len(segments) > duration / 20: # Too many tiny segments
print("π Silence segmentation not optimal, using time-based")
return self.split_audio_by_time(audio_file_path, chunk_duration)
return segments
else:
print("β° Using time-based segmentation")
return self.split_audio_by_time(audio_file_path, chunk_duration)
except Exception as e:
print(f"β Error in segmentation strategy: {e}")
# Ultimate fallback
return self.split_audio_by_time(audio_file_path, chunk_duration)
def split_audio_locally(
self,
audio_file_path: str,
chunk_duration: int = 60,
use_intelligent_segmentation: bool = True
) -> List[Tuple[str, float, float]]:
"""
Split audio file into chunks locally for distributed processing using intelligent segmentation
Args:
audio_file_path: Path to audio file
chunk_duration: Duration of each chunk in seconds
use_intelligent_segmentation: Whether to use intelligent silence-based segmentation
Returns:
List of (chunk_file_path, start_time, end_time) tuples
"""
try:
# Choose segmentation strategy
segments = self.choose_segmentation_strategy(
audio_file_path,
use_intelligent_segmentation=use_intelligent_segmentation,
chunk_duration=chunk_duration
)
if not segments:
print("β No segments generated")
return []
print(f"π΅ Processing {len(segments)} segments using {segments[0].get('segmentation_type', 'time_based')} segmentation")
# Create temporary directory for chunks
temp_dir = tempfile.mkdtemp(prefix="audio_chunks_")
chunks = []
for segment in segments:
start_time = segment["start_time"]
end_time = segment["end_time"]
duration = segment["duration"]
# Create chunk file path
chunk_filename = f"chunk_{segment['chunk_index']:03d}_{start_time:.1f}s-{end_time:.1f}s.wav"
chunk_path = os.path.join(temp_dir, chunk_filename)
# Extract chunk using ffmpeg-python (no subprocess)
try:
(
ffmpeg
.input(audio_file_path, ss=start_time, t=duration)
.output(
chunk_path,
acodec='pcm_s16le',
ar=16000,
ac=1
)
.overwrite_output()
.run(quiet=True, capture_stdout=True, capture_stderr=True)
)
except ffmpeg.Error as e:
print(f"β FFmpeg error for chunk {segment['chunk_index']+1}: {e}")
print(f" stderr: {e.stderr.decode() if e.stderr else 'No stderr'}")
continue
if os.path.exists(chunk_path) and os.path.getsize(chunk_path) > 0:
chunks.append((chunk_path, start_time, end_time))
segmentation_type = segment.get('segmentation_type', 'time_based')
print(f"π¦ Created {segmentation_type} chunk {segment['chunk_index']+1}: {start_time:.1f}s-{end_time:.1f}s")
else:
print(f"β οΈ Failed to create chunk {segment['chunk_index']+1}")
return chunks
except Exception as e:
print(f"β Error splitting audio: {e}")
return []
async def transcribe_chunk_distributed(
self,
chunk_path: str,
start_time: float,
end_time: float,
model_size: str = "turbo",
language: str = None,
enable_speaker_diarization: bool = False,
chunk_endpoint_url: str = None
) -> Dict[str, Any]:
"""
Transcribe a single chunk using Modal distributed endpoint
Args:
chunk_path: Path to audio chunk file
start_time: Start time of chunk in original audio
end_time: End time of chunk in original audio
model_size: Whisper model size
language: Language code
enable_speaker_diarization: Whether to enable speaker diarization
chunk_endpoint_url: URL of chunk transcription endpoint
Returns:
Transcription result for the chunk
"""
try:
# Read and encode chunk file
with open(chunk_path, "rb") as f:
audio_data = f.read()
audio_base64 = base64.b64encode(audio_data).decode('utf-8')
# Prepare request data
request_data = {
"audio_file_data": audio_base64,
"audio_file_name": os.path.basename(chunk_path),
"model_size": model_size,
"language": language,
"output_format": "json", # Use JSON for easier merging
"enable_speaker_diarization": enable_speaker_diarization,
"chunk_start_time": start_time,
"chunk_end_time": end_time
}
# Send request to Modal chunk endpoint with retry mechanism
max_retries = 3
for attempt in range(max_retries):
try:
# Adjust timeout based on whether speaker diarization is enabled
if enable_speaker_diarization:
timeout_config = aiohttp.ClientTimeout(
total=720, # 12 minutes total for speaker diarization
connect=45, # 45 seconds connection timeout
sock_read=300 # 5 minutes read timeout for speaker processing
)
else:
timeout_config = aiohttp.ClientTimeout(
total=480, # 8 minutes total for regular transcription
connect=30, # 30 seconds connection timeout
sock_read=120 # 2 minutes read timeout for regular processing
)
async with aiohttp.ClientSession(timeout=timeout_config) as session:
async with session.post(
chunk_endpoint_url,
json=request_data
) as response:
if response.status == 200:
result = await response.json()
result["chunk_start_time"] = start_time
result["chunk_end_time"] = end_time
result["chunk_file"] = chunk_path
return result
else:
error_text = await response.text()
if attempt < max_retries - 1:
print(f"β οΈ HTTP {response.status} on attempt {attempt + 1}, retrying...")
await asyncio.sleep(2 ** attempt) # Exponential backoff
continue
else:
return {
"processing_status": "failed",
"error_message": f"HTTP {response.status} after {max_retries} attempts: {error_text}",
"chunk_start_time": start_time,
"chunk_end_time": end_time,
"chunk_file": chunk_path
}
except (asyncio.TimeoutError, aiohttp.ClientError) as e:
if attempt < max_retries - 1:
print(f"β οΈ Network error on attempt {attempt + 1}: {e}, retrying...")
await asyncio.sleep(2 ** attempt) # Exponential backoff
continue
else:
return {
"processing_status": "failed",
"error_message": f"Network error after {max_retries} attempts: {e}",
"chunk_start_time": start_time,
"chunk_end_time": end_time,
"chunk_file": chunk_path
}
except Exception as e:
return {
"processing_status": "failed",
"error_message": str(e),
"chunk_start_time": start_time,
"chunk_end_time": end_time,
"chunk_file": chunk_path
}
async def merge_chunk_results(
self,
chunk_results: List[Dict[str, Any]],
output_format: str = "srt",
enable_speaker_diarization: bool = False,
audio_file_path: str = None
) -> Dict[str, Any]:
"""
Merge transcription results from multiple chunks
Args:
chunk_results: List of chunk transcription results
output_format: Output format (srt, txt, json)
enable_speaker_diarization: Whether speaker diarization was enabled
audio_file_path: Path to original audio file (needed for speaker embedding)
Returns:
Merged transcription result
"""
try:
print(f"π Starting merge_chunk_results: {len(chunk_results)} chunks to process")
# Filter successful chunks
successful_chunks = [
chunk for chunk in chunk_results
if chunk.get("processing_status") == "success"
]
failed_chunks = [
chunk for chunk in chunk_results
if chunk.get("processing_status") != "success"
]
print(f"π Chunk processing results: {len(successful_chunks)} successful, {len(failed_chunks)} failed")
if not successful_chunks:
print("β All chunks failed - returning failure result")
return {
"processing_status": "failed",
"error_message": "All chunks failed to process",
"chunks_processed": 0,
"chunks_failed": len(failed_chunks)
}
# Sort chunks by start time
successful_chunks.sort(key=lambda x: x.get("chunk_start_time", 0))
print(f"π Sorted {len(successful_chunks)} successful chunks by start time")
# Apply speaker embedding unification if speaker diarization is enabled
speaker_mapping = {}
if enable_speaker_diarization and audio_file_path:
print(f"π€ Speaker diarization enabled, attempting speaker unification...")
try:
from .speaker_embedding_service import SpeakerIdentificationService, SpeakerEmbeddingService
from ..utils.config import AudioProcessingConfig
print(f"β
Successfully imported speaker embedding services")
# Initialize speaker services
embedding_manager = SpeakerEmbeddingService()
speaker_service = SpeakerIdentificationService(embedding_manager)
print(f"β
Speaker services initialized")
# Unify speakers across chunks using embedding similarity
print("π€ Unifying speakers across chunks using embedding similarity...")
speaker_mapping = await speaker_service.unify_distributed_speakers(
successful_chunks, audio_file_path
)
print(f"β
Speaker unification returned mapping with {len(speaker_mapping)} entries")
if speaker_mapping:
print(f"β
Speaker unification completed: {len(set(speaker_mapping.values()))} unified speakers")
else:
print("β οΈ Speaker unification returned empty mapping")
except Exception as e:
print(f"β οΈ Speaker unification failed: {e}")
print(f" Exception type: {type(e).__name__}")
import traceback
print(f" Traceback: {traceback.format_exc()}")
print("π Continuing with original speaker labels...")
speaker_mapping = {}
else:
if enable_speaker_diarization:
print("β οΈ Speaker diarization enabled but no audio_file_path provided")
if audio_file_path:
print("βΉοΈ Audio file path provided but speaker diarization disabled")
# Merge segments
all_segments = []
total_duration = 0
segment_count = 0
# First pass: collect all segments and mark missing speakers as UNKNOWN
print("π First pass: collecting segments and marking unknown speakers...")
for chunk_idx, chunk in enumerate(successful_chunks):
chunk_start = chunk.get("chunk_start_time", 0)
chunk_segments = chunk.get("segments", [])
for segment in chunk_segments:
# Adjust segment timestamps to global timeline
adjusted_segment = segment.copy()
adjusted_segment["start"] = segment["start"] + chunk_start
adjusted_segment["end"] = segment["end"] + chunk_start
# Mark segments without speaker as UNKNOWN
if "speaker" not in segment or not segment["speaker"]:
adjusted_segment["speaker"] = "UNKNOWN"
adjusted_segment["chunk_id"] = chunk_idx
else:
# Preserve original speaker for embedding-based reassignment
adjusted_segment["original_speaker"] = segment["speaker"]
adjusted_segment["chunk_id"] = chunk_idx
# Temporarily use chunk-local speaker ID for embedding processing
adjusted_segment["speaker"] = f"chunk_{chunk_idx}_{segment['speaker']}"
all_segments.append(adjusted_segment)
segment_count += len(chunk_segments)
chunk_duration = chunk.get("audio_duration", 0)
if chunk_duration > 0:
total_duration = max(total_duration, chunk_start + chunk_duration)
print(f"π Collected {len(all_segments)} segments from {len(successful_chunks)} chunks")
# Second pass: Apply embedding-based speaker unification if enabled
final_speaker_mapping = {}
if enable_speaker_diarization and audio_file_path and speaker_mapping:
print("π€ Second pass: applying embedding-based speaker unification...")
# Create final speaker mapping based on embedding results
for mapping_key, unified_speaker_id in speaker_mapping.items():
final_speaker_mapping[mapping_key] = unified_speaker_id
# Apply the unified speaker mapping to segments
for segment in all_segments:
if segment["speaker"] != "UNKNOWN":
chunk_id = segment["chunk_id"]
original_speaker = segment.get("original_speaker", "")
mapping_key = f"chunk_{chunk_id}_{original_speaker}"
if mapping_key in final_speaker_mapping:
segment["speaker"] = final_speaker_mapping[mapping_key]
print(f"π― Mapped chunk_{chunk_id}_{original_speaker} -> {segment['speaker']}")
else:
# Fallback: create a new speaker ID if not found in mapping
segment["speaker"] = f"SPEAKER_UNMATCHED_{chunk_id}_{original_speaker}"
print(f"β οΈ No mapping found for {mapping_key}, using fallback ID")
print(f"β
Applied speaker unification to segments")
else:
print("βΉοΈ Speaker diarization disabled or no speaker mapping available")
# For segments with speakers but no diarization, use chunk-local naming
for segment in all_segments:
if segment["speaker"] != "UNKNOWN" and segment["speaker"].startswith("chunk_"):
chunk_id = segment["chunk_id"]
original_speaker = segment.get("original_speaker", "")
segment["speaker"] = f"SPEAKER_CHUNK_{chunk_id}_{original_speaker}"
# Third pass: Filter and generate output files
print("π Third pass: generating output files...")
# Separate segments by speaker type
known_speaker_segments = [seg for seg in all_segments if seg["speaker"] != "UNKNOWN"]
unknown_speaker_segments = [seg for seg in all_segments if seg["speaker"] == "UNKNOWN"]
# Only filter UNKNOWN speakers if:
# 1. Speaker diarization is enabled, AND
# 2. There are some known speakers (meaning diarization was successful)
should_filter_unknown = enable_speaker_diarization and len(known_speaker_segments) > 0
if should_filter_unknown:
print(f"π Segment distribution (diarization enabled, filtering UNKNOWN):")
print(f" Known speakers: {len(known_speaker_segments)} segments")
print(f" Unknown speakers: {len(unknown_speaker_segments)} segments (will be filtered)")
# Use only known speaker segments
segments_for_output = known_speaker_segments
else:
# When diarization is disabled OR no speakers were successfully identified,
# use all segments regardless of speaker label
if enable_speaker_diarization:
print(f"π Segment distribution (diarization enabled, but no speakers identified):")
print(f" All segments: {len(all_segments)} segments (no speaker filtering - diarization failed)")
else:
print(f"π Segment distribution (diarization disabled):")
print(f" All segments: {len(all_segments)} segments (no speaker filtering)")
# Use all segments
segments_for_output = all_segments
unknown_speaker_segments = [] # Don't count as filtered if we're not filtering
# Generate output files
output_files = self._generate_output_files(
segments_for_output,
output_format,
should_filter_unknown
)
# Collect speaker information based on filtered segments
speaker_info = self._collect_speaker_information_from_segments(
segments_for_output, enable_speaker_diarization
)
# Determine language (use most common language from chunks)
languages = [chunk.get("language_detected", "unknown") for chunk in successful_chunks]
most_common_language = max(set(languages), key=languages.count) if languages else "unknown"
# Combine text from segments used for output
full_text = " ".join([seg.get("text", "").strip() for seg in segments_for_output if seg.get("text", "").strip()])
print(f"π merge_chunk_results completion summary:")
print(f" Total segments collected: {len(all_segments)}")
print(f" Output segments: {len(segments_for_output)}")
print(f" Unknown speaker segments filtered: {len(unknown_speaker_segments)}")
print(f" Final text length: {len(full_text)} characters")
print(f" Language detected: {most_common_language}")
print(f" Distributed processing flag: True")
return {
"processing_status": "success",
"txt_file_path": output_files.get("txt_file_path"),
"srt_file_path": output_files.get("srt_file_path"),
"audio_duration": total_duration,
"segment_count": len(segments_for_output), # Count segments used for output
"total_segments_collected": len(all_segments), # Total including any filtered segments
"unknown_segments_filtered": len(unknown_speaker_segments), # UNKNOWN segments count (0 if diarization disabled)
"language_detected": most_common_language,
"model_used": successful_chunks[0].get("model_used", "turbo") if successful_chunks else "turbo",
"distributed_processing": True,
"chunks_processed": len(successful_chunks),
"chunks_failed": len(failed_chunks),
"speaker_diarization_enabled": enable_speaker_diarization,
"speaker_embedding_unified": len(speaker_mapping) > 0 if speaker_mapping else False,
"text": full_text, # Add full text for client-side file saving
"segments": segments_for_output, # Add segments for client-side file saving
**speaker_info
}
except Exception as e:
print(f"β Error in merge_chunk_results: {e}")
print(f" Exception type: {type(e).__name__}")
import traceback
print(f" Traceback: {traceback.format_exc()}")
return {
"processing_status": "failed",
"error_message": f"Error merging chunk results: {e}",
"chunks_processed": len(successful_chunks) if 'successful_chunks' in locals() else 0,
"chunks_failed": len(failed_chunks) if 'failed_chunks' in locals() else len(chunk_results)
}
def _generate_output_files(
self,
segments: List[Dict],
output_format: str,
should_filter_unknown: bool
) -> Dict[str, str]:
"""Generate output files from merged segments (filter UNKNOWN speakers only if should_filter_unknown is True)"""
try:
# Create output directory
output_dir = Path(self.cache_dir) / "transcribe"
output_dir.mkdir(parents=True, exist_ok=True)
# Generate timestamp for unique filenames
timestamp = int(time.time())
base_filename = f"distributed_transcription_{timestamp}"
output_files = {}
# Filter segments: only include segments with actual text content
valid_segments = []
for segment in segments:
text = segment.get("text", "").strip()
speaker = segment.get("speaker", "UNKNOWN")
# Skip segments with no text
if not text:
continue
# Only skip UNKNOWN speakers if filtering is enabled
if should_filter_unknown and speaker == "UNKNOWN":
continue
valid_segments.append(segment)
print(f"π Generating output files with {len(valid_segments)} valid segments (filtered from {len(segments)} total)")
# Generate TXT file
txt_path = output_dir / f"{base_filename}.txt"
with open(txt_path, "w", encoding="utf-8") as f:
for segment in valid_segments:
text = segment.get("text", "").strip()
if should_filter_unknown and "speaker" in segment and segment["speaker"] != "UNKNOWN":
f.write(f"[{segment['speaker']}] {text}\n")
else:
f.write(f"{text}\n")
output_files["txt_file_path"] = str(txt_path)
# Generate SRT file if requested
if output_format in ["srt", "both"]:
srt_path = output_dir / f"{base_filename}.srt"
with open(srt_path, "w", encoding="utf-8") as f:
srt_index = 1
for segment in valid_segments:
start_time = self._format_srt_time(segment.get("start", 0))
end_time = self._format_srt_time(segment.get("end", 0))
text = segment.get("text", "").strip()
if should_filter_unknown and "speaker" in segment and segment["speaker"] != "UNKNOWN":
text = f"[{segment['speaker']}] {text}"
f.write(f"{srt_index}\n")
f.write(f"{start_time} --> {end_time}\n")
f.write(f"{text}\n\n")
srt_index += 1
output_files["srt_file_path"] = str(srt_path)
print(f"β
Generated output files: {list(output_files.keys())}")
return output_files
except Exception as e:
print(f"β Error generating output files: {e}")
return {}
def _format_srt_time(self, seconds: float) -> str:
"""Format seconds to SRT time format"""
hours = int(seconds // 3600)
minutes = int((seconds % 3600) // 60)
secs = int(seconds % 60)
millisecs = int((seconds % 1) * 1000)
return f"{hours:02d}:{minutes:02d}:{secs:02d},{millisecs:03d}"
def _collect_speaker_information_from_segments(
self,
segments: List[Dict],
enable_speaker_diarization: bool
) -> Dict[str, Any]:
"""Collect and merge speaker information from segments"""
if not enable_speaker_diarization:
return {}
try:
# Collect all speakers from segments
all_speakers = set()
speaker_summary = {}
for segment in segments:
speaker = segment.get("speaker", "UNKNOWN")
if speaker != "UNKNOWN":
all_speakers.add(speaker)
if speaker not in speaker_summary:
speaker_summary[speaker] = {
"total_duration": 0,
"segment_count": 0
}
# Calculate segment duration from start and end times
segment_duration = segment.get("end", 0) - segment.get("start", 0)
speaker_summary[speaker]["total_duration"] += segment_duration
speaker_summary[speaker]["segment_count"] += 1
return {
"global_speaker_count": len(all_speakers),
"speakers_detected": list(all_speakers),
"speaker_summary": speaker_summary
}
except Exception as e:
print(f"β οΈ Error collecting speaker information: {e}")
print(f" Segment data types: {[type(seg.get('duration', 0)) for seg in segments]}")
return {
"global_speaker_count": 0,
"speakers_detected": [],
"speaker_summary": {}
}
async def transcribe_audio_distributed(
self,
audio_file_path: str,
model_size: str = "turbo",
language: str = None,
output_format: str = "srt",
enable_speaker_diarization: bool = False,
chunk_duration: int = 60,
use_intelligent_segmentation: bool = True,
chunk_endpoint_url: str = None
) -> Dict[str, Any]:
"""
Transcribe audio using distributed processing across multiple Modal containers
Args:
audio_file_path: Path to audio file
model_size: Whisper model size
language: Language code
output_format: Output format
enable_speaker_diarization: Whether to enable speaker diarization
chunk_duration: Duration of each chunk in seconds
use_intelligent_segmentation: Whether to use intelligent segmentation
chunk_endpoint_url: URL of chunk transcription endpoint
Returns:
Transcription result dictionary
"""
temp_files = []
try:
print(f"π Starting distributed transcription for: {audio_file_path}")
print(f"π Using model: {model_size}")
print(f"β‘ Chunk duration: {chunk_duration}s")
# Step 1: Split audio locally into chunks
chunks = self.split_audio_locally(
audio_file_path,
chunk_duration,
use_intelligent_segmentation
)
if not chunks:
return {
"processing_status": "failed",
"error_message": "Failed to split audio into chunks"
}
temp_files.extend([chunk[0] for chunk in chunks])
# Step 2: Process all chunks concurrently (no batching)
print(f"π Processing {len(chunks)} chunks concurrently across multiple containers...")
# Set default chunk endpoint URL if not provided
if not chunk_endpoint_url:
chunk_endpoint_url = "https://richardsucran--transcribe-audio-chunk-endpoint.modal.run"
# Create all tasks simultaneously for maximum concurrency
all_tasks = []
for chunk_idx, (chunk_path, start_time, end_time) in enumerate(chunks):
# Create a coroutine first
coro = self.transcribe_chunk_distributed(
chunk_path=chunk_path,
start_time=start_time,
end_time=end_time,
model_size=model_size,
language=language,
enable_speaker_diarization=enable_speaker_diarization,
chunk_endpoint_url=chunk_endpoint_url
)
# Convert coroutine to Task explicitly
task = asyncio.create_task(coro)
all_tasks.append((chunk_idx, task))
print(f"π€ Launched {len(all_tasks)} concurrent transcription tasks")
# Process results as they complete (optimal resource utilization)
chunk_results = [None] * len(chunks) # Pre-allocate results array
completed_count = 0
failed_count = 0
# Set timeout based on speaker diarization
total_timeout = 1800 if enable_speaker_diarization else 1200 # 30min vs 20min total
print(f"β° Total processing timeout: {total_timeout//60} minutes")
try:
# Use asyncio.wait with return_when=FIRST_COMPLETED for real-time progress
pending_tasks = {task: chunk_idx for chunk_idx, task in all_tasks}
start_time = asyncio.get_event_loop().time()
while pending_tasks:
# Check for timeout
elapsed = asyncio.get_event_loop().time() - start_time
if elapsed > total_timeout:
print(f"β° Total timeout reached ({total_timeout//60} minutes), cancelling remaining tasks...")
for task in pending_tasks.keys():
task.cancel()
break
# Wait for at least one task to complete
remaining_timeout = total_timeout - elapsed
done, pending = await asyncio.wait(
pending_tasks.keys(),
return_when=asyncio.FIRST_COMPLETED,
timeout=min(60, remaining_timeout) # Check every minute
)
# Process completed tasks
for task in done:
chunk_idx = pending_tasks.pop(task)
try:
result = await task
chunk_results[chunk_idx] = result
if result.get("processing_status") == "success":
completed_count += 1
print(f"β
Chunk {chunk_idx + 1}/{len(chunks)} completed successfully")
else:
failed_count += 1
error_msg = result.get("error_message", "Unknown error")
print(f"β Chunk {chunk_idx + 1}/{len(chunks)} failed: {error_msg}")
except Exception as e:
failed_count += 1
chunk_results[chunk_idx] = {
"processing_status": "failed",
"error_message": str(e),
"chunk_start_time": chunks[chunk_idx][1],
"chunk_end_time": chunks[chunk_idx][2],
"chunk_file": chunks[chunk_idx][0]
}
print(f"β Chunk {chunk_idx + 1}/{len(chunks)} exception: {e}")
# Show progress
total_processed = completed_count + failed_count
if total_processed > 0:
print(f"π Progress: {total_processed}/{len(chunks)} chunks processed "
f"({completed_count} β
, {failed_count} β)")
# Handle any remaining cancelled tasks
for task, chunk_idx in pending_tasks.items():
if chunk_results[chunk_idx] is None:
chunk_results[chunk_idx] = {
"processing_status": "failed",
"error_message": "Task cancelled due to timeout",
"chunk_start_time": chunks[chunk_idx][1],
"chunk_end_time": chunks[chunk_idx][2],
"chunk_file": chunks[chunk_idx][0]
}
failed_count += 1
except Exception as e:
print(f"β Error during concurrent processing: {e}")
# Fill in any missing results
for i, result in enumerate(chunk_results):
if result is None:
chunk_results[i] = {
"processing_status": "failed",
"error_message": f"Processing error: {e}",
"chunk_start_time": chunks[i][1],
"chunk_end_time": chunks[i][2],
"chunk_file": chunks[i][0]
}
print(f"π Concurrent processing completed: {completed_count} successful, {failed_count} failed")
# Step 3: Merge results from all chunks
print("π Merging results from all chunks...")
final_result = await self.merge_chunk_results(
chunk_results,
output_format,
enable_speaker_diarization,
audio_file_path
)
print(f"β
Distributed transcription completed successfully")
print(f" Chunks processed: {final_result.get('chunks_processed', 0)}")
print(f" Chunks failed: {final_result.get('chunks_failed', 0)}")
print(f" Total segments: {final_result.get('segment_count', 0)}")
print(f" Duration: {final_result.get('audio_duration', 0):.2f}s")
return final_result
except Exception as e:
return {
"processing_status": "failed",
"error_message": f"Distributed transcription failed: {e}",
"chunks_processed": 0,
"chunks_failed": len(chunks) if 'chunks' in locals() else 0
}
finally:
# Clean up temporary files
for temp_file in temp_files:
try:
if os.path.exists(temp_file):
os.remove(temp_file)
except Exception as e:
print(f"β οΈ Failed to clean up temp file {temp_file}: {e}")
# Clean up temporary directories
for chunk_path, _, _ in chunks if 'chunks' in locals() else []:
try:
temp_dir = os.path.dirname(chunk_path)
if temp_dir.startswith("/tmp/audio_chunks_"):
import shutil
shutil.rmtree(temp_dir, ignore_errors=True)
except Exception as e:
print(f"β οΈ Failed to clean up temp directory: {e}") |