File size: 16,624 Bytes
76f9cd2 |
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 |
"""
Integration tests for Speaker Embedding functionality
Tests the complete workflow of speaker unification in distributed transcription
"""
import pytest
import asyncio
import tempfile
import shutil
import json
import numpy as np
from pathlib import Path
from unittest.mock import Mock, patch, AsyncMock, MagicMock
import torch
from src.services.distributed_transcription_service import DistributedTranscriptionService
from src.services.speaker_embedding_service import SpeakerEmbeddingService, SpeakerIdentificationService
from src.utils.config import AudioProcessingConfig
class TestSpeakerEmbeddingIntegration:
"""Integration tests for speaker embedding with distributed transcription"""
def setup_method(self):
"""Setup test environment"""
self.temp_dir = tempfile.mkdtemp()
self.service = DistributedTranscriptionService(cache_dir=self.temp_dir)
def teardown_method(self):
"""Cleanup test environment"""
shutil.rmtree(self.temp_dir, ignore_errors=True)
@pytest.mark.asyncio
@patch.dict('os.environ', {'HF_TOKEN': 'test_token'})
async def test_merge_chunk_results_with_speaker_unification(self):
"""Test complete speaker unification workflow in merge_chunk_results"""
# Create realistic chunk results with overlapping speakers
chunk_results = [
{
"processing_status": "success",
"chunk_start_time": 0.0,
"audio_duration": 60.0,
"language_detected": "en",
"model_used": "turbo",
"segments": [
{"start": 0.0, "end": 5.0, "text": "Hello everyone", "speaker": "SPEAKER_00"},
{"start": 5.0, "end": 10.0, "text": "How are you?", "speaker": "SPEAKER_01"},
{"start": 10.0, "end": 15.0, "text": "I'm doing well", "speaker": "SPEAKER_00"},
]
},
{
"processing_status": "success",
"chunk_start_time": 60.0,
"audio_duration": 60.0,
"language_detected": "en",
"model_used": "turbo",
"segments": [
{"start": 0.0, "end": 5.0, "text": "Let's continue", "speaker": "SPEAKER_00"}, # Same as chunk 0 SPEAKER_00
{"start": 5.0, "end": 10.0, "text": "Great idea", "speaker": "SPEAKER_01"}, # Same as chunk 0 SPEAKER_01
{"start": 10.0, "end": 15.0, "text": "New person here", "speaker": "SPEAKER_02"}, # New speaker
]
}
]
with patch('src.services.speaker_embedding_service.SpeakerIdentificationService') as mock_service_class:
# Configure the mock speaker service
mock_service = Mock()
mock_service.unify_distributed_speakers = AsyncMock()
# Create a realistic speaker mapping
mock_speaker_mapping = {
"chunk_0_SPEAKER_00": "SPEAKER_GLOBAL_001", # Person A
"chunk_0_SPEAKER_01": "SPEAKER_GLOBAL_002", # Person B
"chunk_1_SPEAKER_00": "SPEAKER_GLOBAL_001", # Person A (unified)
"chunk_1_SPEAKER_01": "SPEAKER_GLOBAL_002", # Person B (unified)
"chunk_1_SPEAKER_02": "SPEAKER_GLOBAL_003", # Person C (new)
}
mock_service.unify_distributed_speakers.return_value = mock_speaker_mapping
mock_service_class.return_value = mock_service
# Test the merge functionality
result = await self.service.merge_chunk_results(
chunk_results=chunk_results,
output_format="srt",
enable_speaker_diarization=True,
audio_file_path="test_audio.wav"
)
# Verify the result
assert result["processing_status"] == "success"
assert result["speaker_diarization_enabled"] is True
assert result["speaker_embedding_unified"] is True
assert result["distributed_processing"] is True
assert result["chunks_processed"] == 2
assert result["chunks_failed"] == 0
# Check speaker statistics
assert result["global_speaker_count"] == 3 # Should be unified to 3 speakers
expected_speakers = {"SPEAKER_GLOBAL_001", "SPEAKER_GLOBAL_002", "SPEAKER_GLOBAL_003"}
assert set(result["speakers_detected"]) == expected_speakers
# Check segments have unified speaker labels
segments = result["segments"]
assert len(segments) == 6 # Total segments across all chunks
# Verify speaker mappings in segments
for segment in segments:
assert "speaker" in segment
assert segment["speaker"] in expected_speakers
@pytest.mark.asyncio
async def test_merge_chunk_results_without_speaker_diarization(self):
"""Test merge_chunk_results when speaker diarization is disabled"""
chunk_results = [
{
"processing_status": "success",
"chunk_start_time": 0.0,
"audio_duration": 60.0,
"language_detected": "en",
"model_used": "turbo",
"segments": [
{"start": 0.0, "end": 5.0, "text": "Hello everyone"},
{"start": 5.0, "end": 10.0, "text": "How are you?"},
]
}
]
result = await self.service.merge_chunk_results(
chunk_results=chunk_results,
output_format="srt",
enable_speaker_diarization=False,
audio_file_path="test_audio.wav"
)
# Should not perform speaker unification
assert result["processing_status"] == "success"
assert result["speaker_diarization_enabled"] is False
assert "speaker_embedding_unified" not in result or result["speaker_embedding_unified"] is False
# Note: global_speaker_count may not be present when speaker diarization is disabled
@pytest.mark.asyncio
async def test_merge_chunk_results_speaker_unification_failure(self):
"""Test merge_chunk_results when speaker unification fails"""
chunk_results = [
{
"processing_status": "success",
"chunk_start_time": 0.0,
"audio_duration": 60.0,
"language_detected": "en",
"model_used": "turbo",
"segments": [
{"start": 0.0, "end": 5.0, "text": "Hello", "speaker": "SPEAKER_00"},
]
}
]
with patch('src.services.speaker_embedding_service.SpeakerIdentificationService') as mock_service_class:
# Make the speaker service throw an exception
mock_service = Mock()
mock_service.unify_distributed_speakers = AsyncMock(side_effect=Exception("Model not available"))
mock_service_class.return_value = mock_service
result = await self.service.merge_chunk_results(
chunk_results=chunk_results,
output_format="srt",
enable_speaker_diarization=True,
audio_file_path="test_audio.wav"
)
# Should continue with original speaker labels
assert result["processing_status"] == "success"
assert result["speaker_diarization_enabled"] is True
assert "speaker_embedding_unified" not in result or result["speaker_embedding_unified"] is False
# Should have chunk-aware speaker labels when unification fails
segments = result["segments"]
assert len(segments) == 1
assert segments[0]["speaker"] == "SPEAKER_CHUNK_0_SPEAKER_00" # Chunk-aware label when unification fails
@pytest.mark.asyncio
async def test_merge_chunk_results_unknown_speaker_filtering(self):
"""Test that UNKNOWN speakers are properly filtered from output"""
service = DistributedTranscriptionService()
# Mock chunk results with mixed speaker types
chunk_results = [
{
"processing_status": "success",
"chunk_start_time": 0.0,
"chunk_end_time": 30.0,
"segments": [
{
"start": 0.0,
"end": 2.0,
"text": "Hello world",
"speaker": "SPEAKER_00"
},
{
"start": 2.0,
"end": 4.0,
"text": "This has no speaker",
# No speaker field - should become UNKNOWN
},
{
"start": 4.0,
"end": 6.0,
"text": "Another good segment",
"speaker": "SPEAKER_01"
}
]
},
{
"processing_status": "success",
"chunk_start_time": 30.0,
"chunk_end_time": 60.0,
"segments": [
{
"start": 0.0,
"end": 2.0,
"text": "", # Empty text - should be filtered
"speaker": "SPEAKER_00"
},
{
"start": 2.0,
"end": 4.0,
"text": "Good segment from chunk 2",
"speaker": "SPEAKER_00"
}
]
}
]
# Mock speaker embedding service (not available)
with patch('src.services.distributed_transcription_service.SpeakerIdentificationService', create=True) as mock_service_class:
with patch('src.services.distributed_transcription_service.SpeakerEmbeddingService', create=True) as mock_manager_class:
result = await service.merge_chunk_results(
chunk_results=chunk_results,
output_format="srt",
enable_speaker_diarization=False, # Disable to focus on filtering logic
audio_file_path="test.wav"
)
# Verify basic result structure
assert result["processing_status"] == "success"
assert result["chunks_processed"] == 2
assert result["chunks_failed"] == 0
# Verify filtering statistics
assert "total_segments_collected" in result
assert "unknown_segments_filtered" in result
assert result["total_segments_collected"] == 5 # Total segments before filtering (4 with data + 1 empty)
assert result["unknown_segments_filtered"] == 1 # One segment with no speaker
assert result["segment_count"] == 4 # Known speaker segments after first filtering (including empty text)
# Verify segments content (this will be the final filtered segments)
segments = result["segments"]
assert len(segments) == 4 # Known speaker segments (including empty text ones)
# Check that all segments have speakers (but may have empty text)
for segment in segments:
assert "speaker" in segment
assert segment["speaker"] != "UNKNOWN"
# Note: segments array includes all known-speaker segments,
# but full_text and output files filter empty text
# Verify text content (should not include UNKNOWN segment)
full_text = result["text"]
assert "This has no speaker" not in full_text # UNKNOWN segment filtered
assert "Hello world" in full_text
assert "Another good segment" in full_text
assert "Good segment from chunk 2" in full_text
class TestSpeakerEmbeddingWorkflow:
"""Test complete workflow scenarios"""
def setup_method(self):
"""Setup test environment"""
self.temp_dir = tempfile.mkdtemp()
def teardown_method(self):
"""Cleanup test environment"""
shutil.rmtree(self.temp_dir, ignore_errors=True)
@pytest.mark.asyncio
@patch.dict('os.environ', {'HF_TOKEN': 'test_token'})
async def test_end_to_end_speaker_unification(self):
"""Test complete end-to-end speaker unification workflow"""
# Setup services
embedding_service = SpeakerEmbeddingService(
storage_path=str(Path(self.temp_dir) / "speakers.json")
)
speaker_service = SpeakerIdentificationService(embedding_service)
# Mock the models
speaker_service.embedding_model = Mock()
# Create test chunk results representing a conversation between 2 people
# but each chunk detects them as different local speakers
chunk_results = [
{
"processing_status": "success",
"chunk_start_time": 0,
"segments": [
{"start": 0, "end": 3, "text": "Hi there", "speaker": "SPEAKER_00"},
{"start": 3, "end": 6, "text": "Hello", "speaker": "SPEAKER_01"},
]
},
{
"processing_status": "success",
"chunk_start_time": 60,
"segments": [
{"start": 0, "end": 3, "text": "How are things?", "speaker": "SPEAKER_00"}, # Same person as chunk 0 SPEAKER_00
{"start": 3, "end": 6, "text": "Going well", "speaker": "SPEAKER_01"}, # Same person as chunk 0 SPEAKER_01
]
}
]
# Create consistent embeddings for same speakers
person_a_base = np.zeros(512)
person_a_base[0] = 1.0 # Person A concentrated at index 0
person_b_base = np.zeros(512)
person_b_base[256] = 1.0 # Person B concentrated at index 256
def mock_crop_side_effect(waveform, segment):
# Return consistent embeddings for same speakers across chunks
segment_start = float(segment.start)
if segment_start < 3 or (segment_start >= 60 and segment_start < 63):
# Person A in both chunks
return torch.tensor(person_a_base + np.random.normal(0, 0.005, 512))
else:
# Person B in both chunks
return torch.tensor(person_b_base + np.random.normal(0, 0.005, 512))
mock_inference = Mock()
mock_inference.crop.side_effect = mock_crop_side_effect
mock_waveform = torch.rand(1, 96000) # 6 seconds at 16kHz
with patch('torchaudio.load', return_value=(mock_waveform, 16000)), \
patch('pyannote.audio.core.inference.Inference', return_value=mock_inference):
# Perform speaker unification
mapping = await speaker_service.unify_distributed_speakers(
chunk_results, "test_audio.wav"
)
# Verify mapping results
assert len(mapping) == 4 # 2 speakers × 2 chunks
# Same speakers should map to same global IDs
chunk_0_speaker_00 = mapping["chunk_0_SPEAKER_00"]
chunk_1_speaker_00 = mapping["chunk_1_SPEAKER_00"]
chunk_0_speaker_01 = mapping["chunk_0_SPEAKER_01"]
chunk_1_speaker_01 = mapping["chunk_1_SPEAKER_01"]
# Verify unification worked
assert chunk_0_speaker_00 == chunk_1_speaker_00 # Same person A
assert chunk_0_speaker_01 == chunk_1_speaker_01 # Same person B
assert chunk_0_speaker_00 != chunk_0_speaker_01 # Different people
# Verify global speaker IDs are properly formatted
assert chunk_0_speaker_00.startswith("SPEAKER_GLOBAL_")
assert chunk_0_speaker_01.startswith("SPEAKER_GLOBAL_")
if __name__ == "__main__":
pytest.main([__file__, "-v"]) |