Spaces:
Sleeping
Sleeping
File size: 10,437 Bytes
8c85b97 | 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 | # core/transcription_service.py
import whisperx
import gradio as gr
import gc
import time
import os
from datetime import datetime
import warnings
# Suppress specific pyannote/pytorch warning about degrees of freedom
warnings.filterwarnings("ignore", message="std\(\): degrees of freedom is <= 0")
from src.config.settings import Config
# CORRECT WAY: Import DiarizationPipeline at point of use
from whisperx.diarize import DiarizationPipeline
class TranscriptionService:
def __init__(self):
self.config = Config
self.models_loaded = False
self.whisper_model = None
self.diarize_model = None
self.batch_size = 16
def load_models(self):
"""Load AI models once - use pre-loaded models from init"""
if not self.models_loaded:
print("📥 Loading transcription models...")
# Use the model from config instead of hardcoding
self.whisper_model = whisperx.load_model(
self.config.WHISPER_MODEL,
self.config.DEVICE,
compute_type=self.config.COMPUTE_TYPE,
language="en"
)
self.diarize_model = DiarizationPipeline(
use_auth_token=self.config.HUGGINGFACE_TOKEN,
device=self.config.DEVICE
)
self.models_loaded = True
print("✅ Models loaded successfully")
def transcribe_video(self, video_file_path, progress_callback=None):
"""Clean transcription pipeline without Gradio dependencies.
Added optional progress callback"""
try:
if not self.models_loaded:
self.load_models()
start_time = time.time()
print(f"🎬 Processing video: {os.path.basename(video_file_path)}")
# ======================
# STEP 1: Load Audio from Video
# ======================
if progress_callback:
print(f"DEBUG: Calling progress callback 0.1. Type: {type(progress_callback)}")
try:
progress_callback(0.1, desc="🎬 Loading audio from video...")
print("DEBUG: Progress callback 0.1 called successfully")
except Exception as e:
print(f"DEBUG: Error calling progress callback: {e}")
time.sleep(0.5)
print("1️⃣ Loading audio directly from video...")
audio = whisperx.load_audio(video_file_path)
print(f"✅ Audio loaded: {len(audio)} samples")
# ======================
# STEP 2: Transcribe with Whisper
# ======================
print("2️⃣ Loading Whisper model...")
if progress_callback:
progress_callback(0.3, desc="🤖 Loading Whisper model...")
time.sleep(0.5)
if progress_callback:
progress_callback(0.4, desc="📝 Transcribing audio...")
time.sleep(0.5)
print("3️⃣ Transcribing audio...")
result = self.whisper_model.transcribe(audio, batch_size=self.batch_size)
detected_language = result['language'] # Save language before it gets lost
print(f"✅ Transcription complete ({detected_language} detected)")
# ======================
# STEP 3: Align Timestamps
# ======================
if progress_callback:
progress_callback(0.5, desc="⏱️ Aligning timestamps...")
time.sleep(0.5)
print("4️⃣ Aligning word-level timestamps...")
# Load the alignment model and its metadata from whisperx for word-level timestamp alignment.
model_a, metadata = whisperx.load_align_model(
language_code=detected_language,
device=self.config.DEVICE
)
result = whisperx.align(
result["segments"],
model_a,
metadata,
audio,
self.config.DEVICE,
return_char_alignments=False
)
# Restore language to result dict after alignment
result["language"] = detected_language
print("✅ Timestamps aligned")
# ======================
# STEP 4: Speaker Diarization - CORRECT IMPORT
# ======================
if progress_callback:
progress_callback(0.7, desc="👥 Identifying speakers...")
time.sleep(0.5)
print("5️⃣ Loading speaker diarization model...")
diarize_segments = self.diarize_model(audio)
# ======================
# STEP 5: Assign speakers
# ======================
#
if progress_callback:
progress_callback(0.9, desc="🔗 Assigning speakers to text...")
time.sleep(0.5)
result = whisperx.assign_word_speakers(diarize_segments, result)
print("6️⃣ Assigning speakers to transcript...")
print("🔗 Assigning speakers to text...")
result = whisperx.assign_word_speakers(diarize_segments, result)
print("✅ Speaker assignment complete")
if progress_callback:
progress_callback(1.0, desc="✅ Complete!")
time.sleep(0.5)
# ======================
# STEP 6: Format results
# ======================
processing_time = time.time() - start_time
transcription = self._format_results(result, video_file_path)
timing_info = self._get_timing_info(result, processing_time, video_file_path)
return {
"success": True,
"transcription": transcription,
"timing_info": timing_info,
"raw_data": result, # Keep for potential storage
"processing_time": processing_time,
"speakers_count": len(set(seg.get("speaker", "UNKNOWN") for seg in result["segments"]))
}
except Exception as e:
error_msg = f"Transcription failed: {str(e)}"
print(f"❌ ERROR: {error_msg}")
return {
"success": False,
"error": error_msg
}
def _format_results(self, result, video_file_path):
"""Format transcription with speaker labels and comprehensive meeting metadata"""
if not result["segments"]:
return "No transcription segments found"
# Extract meeting metadata
segments = result["segments"]
speakers = set(segment.get("speaker", "UNKNOWN") for segment in segments)
total_duration = segments[-1]["end"] if segments else 0
language = result.get("language", "unknown")
# Calculate statistics
total_words = sum(len(seg.get("text", "").split()) for seg in segments)
avg_segment_length = total_words / len(segments) if segments else 0
# Build header with meeting context
output = "# 🎯 Meeting Transcription\n\n"
output += "## 📋 Meeting Information\n\n"
output += f"**📁 File:** `{os.path.basename(video_file_path)}`\n"
output += f"**📅 Date:** {datetime.now().strftime('%Y-%m-%d %H:%M')}\n"
output += f"**⏱️ Duration:** {self._format_timestamp(total_duration)}\n"
output += f"**👥 Speakers:** {len(speakers)}\n"
output += f"**🌐 Language:** {language.upper()}\n"
output += f"**🤖 Model:** {self.config.WHISPER_MODEL}\n\n"
output += "---\n\n"
output += "## 💬 Transcript\n\n"
# Add transcript content
current_speaker = None
for segment in segments:
speaker = segment.get("speaker", "UNKNOWN")
start_time = self._format_timestamp(segment["start"])
if speaker != current_speaker:
output += f"\n**👤 {speaker}:**\n"
current_speaker = speaker
output += f"[{start_time}] {segment['text'].strip()}\n"
# Add comprehensive footer
output += "\n---\n\n"
output += "## 📊 Transcript Statistics\n\n"
output += f"**Total Segments:** {len(segments)}\n"
output += f"**Total Words:** {total_words:,}\n"
output += f"**Avg Words/Segment:** {avg_segment_length:.1f}\n"
output += f"**Unique Speakers:** {len(speakers)}\n"
output += f"**Speaker IDs:** {', '.join(sorted(speakers))}\n"
return output
def _get_timing_info(self, result, processing_time, video_file_path):
"""Generate timing information"""
if not result["segments"]:
return "No timing information available"
total_duration = result["segments"][-1]["end"]
speed_ratio = total_duration / processing_time if processing_time > 0 else 0
video_name = os.path.basename(video_file_path)
return f"""
## ⏱️ Processing Statistics
**File:** {video_name}
\n**Duration:** {self._format_timestamp(total_duration)}
**Processing Time:** {processing_time:.1f}s
\n**Speed:** {speed_ratio:.1f}x ({'Faster' if speed_ratio > 1 else 'Slower'} than real-time)
**Completed:** {datetime.now().strftime("%H:%M:%S")}
"""
def _format_timestamp(self, seconds):
"""Convert seconds to MM:SS format"""
if seconds is None:
return "00:00"
minutes = int(seconds // 60)
seconds = int(seconds % 60)
return f"{minutes:02d}:{seconds:02d}" |