| 1 | import asyncio |
| 2 | import base64 |
| 3 | import os |
| 4 | import unittest |
| 5 | import sys |
| 6 | import tempfile |
| 7 | import time |
| 8 | from datetime import timedelta |
| 9 | from pathlib import Path |
| 10 | from types import SimpleNamespace |
| 11 | from unittest.mock import patch |
| 12 | |
| 13 | # add project root to python path |
| 14 | sys.path.insert(0, str(Path(__file__).parent.parent.parent)) |
| 15 | |
| 16 | from app.utils import utils |
| 17 | from app.services import voice as vs |
| 18 | from app.services import task as task_service |
| 19 | from pydub import AudioSegment |
| 20 | |
| 21 | temp_dir = utils.storage_dir("temp") |
| 22 | |
| 23 | text_en = """ |
| 24 | What is the meaning of life? |
| 25 | This question has puzzled philosophers, scientists, and thinkers of all kinds for centuries. |
| 26 | Throughout history, various cultures and individuals have come up with their interpretations and beliefs around the purpose of life. |
| 27 | Some say it's to seek happiness and self-fulfillment, while others believe it's about contributing to the welfare of others and making a positive impact in the world. |
| 28 | Despite the myriad of perspectives, one thing remains clear: the meaning of life is a deeply personal concept that varies from one person to another. |
| 29 | It's an existential inquiry that encourages us to reflect on our values, desires, and the essence of our existence. |
| 30 | """ |
| 31 | |
| 32 | text_zh = """ |
| 33 | 预计未来3天深圳冷空气活动频繁，未来两天持续阴天有小雨，出门带好雨具； |
| 34 | 10-11日持续阴天有小雨，日温差小，气温在13-17℃之间，体感阴凉； |
| 35 | 12日天气短暂好转，早晚清凉； |
| 36 | """ |
| 37 | |
| 38 | voice_rate=1.0 |
| 39 | voice_volume=1.0 |
| 40 | RUN_INTEGRATION_TESTS = os.environ.get("MPT_RUN_INTEGRATION_TESTS", "").lower() in { |
| 41 | "1", |
| 42 | "true", |
| 43 | "yes", |
| 44 | } |
| 45 | |
| 46 | class TestVoiceService(unittest.TestCase): |
| 47 | def setUp(self): |
| 48 | self.loop = asyncio.new_event_loop() |
| 49 | asyncio.set_event_loop(self.loop) |
| 50 | |
| 51 | def tearDown(self): |
| 52 | self.loop.close() |
| 53 | |
| 54 | def test_get_all_azure_voices(self): |
| 55 | voices = vs.get_all_azure_voices() |
| 56 | # 数据已从内联字符串迁移到 azure_voices.json，确保仍能完整加载 |
| 57 | self.assertEqual(len(voices), 331) |
| 58 | # 结果应为 "Name-Gender" 格式且已排序 |
| 59 | self.assertEqual(voices, sorted(voices)) |
| 60 | for v in voices: |
| 61 | self.assertTrue(v.endswith("-Male") or v.endswith("-Female")) |
| 62 | |
| 63 | def test_get_all_azure_voices_filtered(self): |
| 64 | filtered = vs.get_all_azure_voices(filter_locals=["zh-CN", "en-US"]) |
| 65 | self.assertTrue(len(filtered) > 0) |
| 66 | self.assertTrue( |
| 67 | all(v.startswith(("zh-CN", "en-US")) for v in filtered) |
| 68 | ) |
| 69 | |
| 70 | def test_no_voice_tts_generates_silent_audio_and_subtitle_timeline(self): |
| 71 | """ |
| 72 | 无配音模式不调用任何外部 TTS provider，只生成静音音频作为时间轴占位。 |
| 73 | 这里 mock FFmpeg，验证请求参数、输出文件和 legacy 字幕结构都符合后续 |
| 74 | 视频合成链路的预期。 |
| 75 | """ |
| 76 | |
| 77 | def fake_run(command, capture_output, text, check): |
| 78 | self.assertEqual(command[0], "/tmp/fake-ffmpeg") |
| 79 | self.assertIn("anullsrc=r=44100:cl=mono", command) |
| 80 | Path(command[-1]).write_bytes(b"fake-silent-mp3") |
| 81 | return SimpleNamespace(returncode=0, stdout="", stderr="") |
| 82 | |
| 83 | with tempfile.TemporaryDirectory() as tmp_dir, patch.object( |
| 84 | vs.utils, |
| 85 | "get_ffmpeg_binary", |
| 86 | return_value="/tmp/fake-ffmpeg", |
| 87 | ), patch.object(vs.subprocess, "run", side_effect=fake_run): |
| 88 | voice_file = str(Path(tmp_dir) / "silent.mp3") |
| 89 | sub_maker = vs.tts( |
| 90 | text="第一句话。Second sentence.", |
| 91 | voice_name=vs.NO_VOICE_NAME, |
| 92 | voice_rate=1.0, |
| 93 | voice_file=voice_file, |
| 94 | ) |
| 95 | |
| 96 | self.assertEqual(Path(voice_file).read_bytes(), b"fake-silent-mp3") |
| 97 | |
| 98 | self.assertIsNotNone(sub_maker) |
| 99 | self.assertEqual(getattr(sub_maker, "subs", []), ["第一句话", "Second sentence"]) |
| 100 | self.assertEqual(len(getattr(sub_maker, "offset", [])), 2) |
| 101 | self.assertGreater(vs.get_audio_duration(sub_maker), 0) |
| 102 | |
| 103 | def test_no_voice_alias_none_is_supported_temporarily(self): |
| 104 | """ |
| 105 | 兼容 PR #981 曾使用过的 none sentinel，避免少量直接调用 API 的用户 |
| 106 | 升级后立即失效。新 UI 和新代码仍统一使用 no-voice。 |
| 107 | """ |
| 108 | self.assertTrue(vs.is_no_voice("none")) |
| 109 | self.assertTrue(vs.is_no_voice(vs.NO_VOICE_NAME)) |
| 110 | self.assertFalse(vs.is_no_voice("")) |
| 111 | |
| 112 | def test_no_voice_duration_estimates_non_ascii_languages(self): |
| 113 | """ |
| 114 | 无配音没有真实 TTS 音频，只能根据脚本文字估算阅读时间。俄语、阿拉伯语、 |
| 115 | 日文假名、韩文等非 ASCII 文本也必须参与估算，不能都落到最短 3 秒。 |
| 116 | """ |
| 117 | russian_text = ( |
| 118 | "Это длинный тестовый сценарий без озвучки. " |
| 119 | "Он должен получить достаточно времени для чтения субтитров." |
| 120 | ) |
| 121 | arabic_text = "هذا اختبار طويل بدون تعليق صوتي، ويجب أن يحصل على وقت كاف لقراءة الترجمة." |
| 122 | |
| 123 | self.assertGreater(vs.estimate_no_voice_duration(russian_text), 8.0) |
| 124 | self.assertGreater(vs.estimate_no_voice_duration(arabic_text), 8.0) |
| 125 | |
| 126 | def test_generate_silent_audio_rejects_missing_output_file(self): |
| 127 | """ |
| 128 | 即使 FFmpeg 进程返回成功，也要确认输出文件真实存在且非空。这样可以把 |
| 129 | 异常收敛在 TTS 阶段，而不是拖到后续视频合成阶段才暴露。 |
| 130 | """ |
| 131 | with tempfile.TemporaryDirectory() as tmp_dir, patch.object( |
| 132 | vs.utils, |
| 133 | "get_ffmpeg_binary", |
| 134 | return_value="/tmp/fake-ffmpeg", |
| 135 | ), patch.object( |
| 136 | vs.subprocess, |
| 137 | "run", |
| 138 | return_value=SimpleNamespace(returncode=0, stdout="", stderr=""), |
| 139 | ): |
| 140 | voice_file = str(Path(tmp_dir) / "missing-silent.mp3") |
| 141 | |
| 142 | self.assertFalse(vs.generate_silent_audio(3.0, voice_file)) |
| 143 | |
| 144 | def test_empty_voice_name_does_not_enable_no_voice_mode(self): |
| 145 | """ |
| 146 | 空 voice 通常意味着配置缺失或接口参数错误，不能自动切到无配音模式。 |
| 147 | 否则用户填错 TTS 配置时也会得到一个“成功”的静音视频，定位成本更高。 |
| 148 | """ |
| 149 | sentinel = object() |
| 150 | |
| 151 | with patch.object(vs, "azure_tts_v1", return_value=sentinel) as azure_tts_v1: |
| 152 | result = vs.tts( |
| 153 | text="empty voice should still use the default TTS path", |
| 154 | voice_name="", |
| 155 | voice_rate=1.0, |
| 156 | voice_file="/tmp/empty-voice.mp3", |
| 157 | ) |
| 158 | |
| 159 | self.assertIs(result, sentinel) |
| 160 | azure_tts_v1.assert_called_once() |
| 161 | |
| 162 | @unittest.skipUnless( |
| 163 | RUN_INTEGRATION_TESTS, |
| 164 | "MPT_RUN_INTEGRATION_TESTS not set", |
| 165 | ) |
| 166 | def test_siliconflow(self): |
| 167 | # SiliconFlow 的 API Key 存在 [siliconflow].api_key 中，运行时代码也是从 |
| 168 | # config.siliconflow 读取；这里必须使用同一配置源，避免正确配置凭据时 |
| 169 | # 测试仍然被误跳过。 |
| 170 | if not vs.config.siliconflow.get("api_key"): |
| 171 | self.skipTest("siliconflow_api_key is not configured") |
| 172 | |
| 173 | voice_name = "siliconflow:FunAudioLLM/CosyVoice2-0.5B:alex-Male" |
| 174 | voice_name = vs.parse_voice_name(voice_name) |
| 175 | |
| 176 | async def _do(): |
| 177 | parts = voice_name.split(":") |
| 178 | if len(parts) >= 3: |
| 179 | model = parts[1] |
| 180 | # 移除性别后缀，例如 "alex-Male" -> "alex" |
| 181 | voice_with_gender = parts[2] |
| 182 | voice = voice_with_gender.split("-")[0] |
| 183 | # 构建完整的voice参数，格式为 "model:voice" |
| 184 | full_voice = f"{model}:{voice}" |
| 185 | voice_file = f"{temp_dir}/tts-siliconflow-{voice}.mp3" |
| 186 | subtitle_file = f"{temp_dir}/tts-siliconflow-{voice}.srt" |
| 187 | sub_maker = vs.siliconflow_tts( |
| 188 | text=text_zh, model=model, voice=full_voice, voice_file=voice_file, voice_rate=voice_rate, voice_volume=voice_volume |
| 189 | ) |
| 190 | if not sub_maker: |
| 191 | self.fail("siliconflow tts failed") |
| 192 | vs.create_subtitle(sub_maker=sub_maker, text=text_zh, subtitle_file=subtitle_file) |
| 193 | audio_duration = vs.get_audio_duration(sub_maker) |
| 194 | print(f"voice: {voice_name}, audio duration: {audio_duration}s") |
| 195 | else: |
| 196 | self.fail("siliconflow invalid voice name") |
| 197 | |
| 198 | self.loop.run_until_complete(_do()) |
| 199 | |
| 200 | @unittest.skipUnless( |
| 201 | RUN_INTEGRATION_TESTS, |
| 202 | "MPT_RUN_INTEGRATION_TESTS not set", |
| 203 | ) |
| 204 | def test_azure_tts_v1(self): |
| 205 | voice_name = "zh-CN-XiaoyiNeural-Female" |
| 206 | voice_name = vs.parse_voice_name(voice_name) |
| 207 | print(voice_name) |
| 208 | |
| 209 | voice_file = f"{temp_dir}/tts-azure-v1-{voice_name}.mp3" |
| 210 | subtitle_file = f"{temp_dir}/tts-azure-v1-{voice_name}.srt" |
| 211 | sub_maker = vs.azure_tts_v1( |
| 212 | text=text_zh, voice_name=voice_name, voice_file=voice_file, voice_rate=voice_rate |
| 213 | ) |
| 214 | if not sub_maker: |
| 215 | self.fail("azure tts v1 failed") |
| 216 | vs.create_subtitle(sub_maker=sub_maker, text=text_zh, subtitle_file=subtitle_file) |
| 217 | audio_duration = vs.get_audio_duration(sub_maker) |
| 218 | print(f"voice: {voice_name}, audio duration: {audio_duration}s") |
| 219 | |
| 220 | def test_azure_tts_v1_supports_legacy_edge_tts_without_boundary(self): |
| 221 | """ |
| 222 | 验证 Azure TTS V1 在旧版 edge_tts 依赖残留时仍可继续工作。 |
| 223 | |
| 224 | 这个回归场景对应 Windows 便携包更新失败后，现场环境还停留在旧版 |
| 225 | edge_tts 的情况： |
| 226 | 1. `Communicate.__init__()` 不接受 `boundary` |
| 227 | 2. 只有异步 `stream()`，没有 `stream_sync()` |
| 228 | """ |
| 229 | |
| 230 | class _LegacyCommunicate: |
| 231 | def __init__(self, text, voice, rate="+0%"): |
| 232 | self.text = text |
| 233 | self.voice = voice |
| 234 | self.rate = rate |
| 235 | |
| 236 | async def stream(self): |
| 237 | yield {"type": "audio", "data": b"legacy-audio"} |
| 238 | yield { |
| 239 | "type": "WordBoundary", |
| 240 | "offset": 0, |
| 241 | "duration": 10000000, |
| 242 | "text": "legacy", |
| 243 | } |
| 244 | |
| 245 | class _FakeSubMaker: |
| 246 | def __init__(self): |
| 247 | self.events = [] |
| 248 | |
| 249 | def feed(self, chunk): |
| 250 | self.events.append(chunk) |
| 251 | |
| 252 | def get_srt(self): |
| 253 | if not self.events: |
| 254 | return "" |
| 255 | return "1\n00:00:00,000 --> 00:00:01,000\nlegacy\n" |
| 256 | |
| 257 | with tempfile.TemporaryDirectory() as tmp_dir, patch.object( |
| 258 | vs.edge_tts, "Communicate", _LegacyCommunicate |
| 259 | ), patch.object(vs.edge_tts, "SubMaker", _FakeSubMaker): |
| 260 | voice_file = str(Path(tmp_dir) / "legacy-edge-tts.mp3") |
| 261 | sub_maker = vs.azure_tts_v1( |
| 262 | text="legacy edge tts compatibility", |
| 263 | voice_name="zh-CN-XiaoyiNeural-Female", |
| 264 | voice_file=voice_file, |
| 265 | voice_rate=1.0, |
| 266 | ) |
| 267 | |
| 268 | self.assertIsNotNone(sub_maker) |
| 269 | self.assertEqual(Path(voice_file).read_bytes(), b"legacy-audio") |
| 270 | self.assertEqual(len(sub_maker.events), 1) |
| 271 | self.assertEqual(sub_maker.events[0]["type"], "WordBoundary") |
| 272 | |
| 273 | def test_azure_tts_v1_times_out_hanging_stream_sync(self): |
| 274 | """ |
| 275 | 验证 Azure TTS V1 在 edge_tts 同步流卡住时能够快速失败。 |
| 276 | |
| 277 | 真实现场里，网络异常、服务端限流、voice 语言与文本不匹配时， |
| 278 | `stream_sync()` 可能长时间不返回，导致 WebUI 任务只停在 |
| 279 | `start, voice name...`。这里用阻塞的 fake stream 复现该场景， |
| 280 | 确认超时保护会让函数结束并返回 None。 |
| 281 | """ |
| 282 | |
| 283 | class _HangingCommunicate: |
| 284 | def __init__(self, text, voice, rate="+0%", boundary=None): |
| 285 | self.text = text |
| 286 | self.voice = voice |
| 287 | self.rate = rate |
| 288 | self.boundary = boundary |
| 289 | |
| 290 | def stream_sync(self): |
| 291 | time.sleep(10) |
| 292 | yield {"type": "audio", "data": b"unreachable"} |
| 293 | |
| 294 | class _FakeSubMaker: |
| 295 | def feed(self, chunk): |
| 296 | return None |
| 297 | |
| 298 | def get_srt(self): |
| 299 | return "" |
| 300 | |
| 301 | with tempfile.TemporaryDirectory() as tmp_dir, patch.object( |
| 302 | vs.edge_tts, "Communicate", _HangingCommunicate |
| 303 | ), patch.object(vs.edge_tts, "SubMaker", _FakeSubMaker), patch.object( |
| 304 | vs.config, |
| 305 | "app", |
| 306 | dict(vs.config.app, edge_tts_timeout=0.05), |
| 307 | ): |
| 308 | voice_file = Path(tmp_dir) / "hanging-edge-tts.mp3" |
| 309 | started_at = time.monotonic() |
| 310 | sub_maker = vs.azure_tts_v1( |
| 311 | text="帮我生成一个花开花落的视频", |
| 312 | voice_name="en-AU-NatashaNeural-Female", |
| 313 | voice_file=str(voice_file), |
| 314 | voice_rate=1.0, |
| 315 | ) |
| 316 | elapsed = time.monotonic() - started_at |
| 317 | self.assertFalse(voice_file.exists()) |
| 318 | |
| 319 | self.assertIsNone(sub_maker) |
| 320 | self.assertLess(elapsed, 2) |
| 321 | |
| 322 | @unittest.skipUnless( |
| 323 | RUN_INTEGRATION_TESTS, |
| 324 | "MPT_RUN_INTEGRATION_TESTS not set", |
| 325 | ) |
| 326 | def test_azure_tts_v2(self): |
| 327 | if not vs.config.azure.get("speech_key") or not vs.config.azure.get("speech_region"): |
| 328 | self.skipTest("Azure speech key or region is not configured") |
| 329 | |
| 330 | voice_name = "zh-CN-XiaoxiaoMultilingualNeural-V2-Female" |
| 331 | voice_name = vs.parse_voice_name(voice_name) |
| 332 | print(voice_name) |
| 333 | |
| 334 | async def _do(): |
| 335 | voice_file = f"{temp_dir}/tts-azure-v2-{voice_name}.mp3" |
| 336 | subtitle_file = f"{temp_dir}/tts-azure-v2-{voice_name}.srt" |
| 337 | sub_maker = vs.azure_tts_v2( |
| 338 | text=text_zh, voice_name=voice_name, voice_file=voice_file |
| 339 | ) |
| 340 | if not sub_maker: |
| 341 | self.fail("azure tts v2 failed") |
| 342 | vs.create_subtitle(sub_maker=sub_maker, text=text_zh, subtitle_file=subtitle_file) |
| 343 | audio_duration = vs.get_audio_duration(sub_maker) |
| 344 | print(f"voice: {voice_name}, audio duration: {audio_duration}s") |
| 345 | |
| 346 | self.loop.run_until_complete(_do()) |
| 347 | |
| 348 | def test_gemini_tts_uses_legacy_submaker_fields(self): |
| 349 | """ |
| 350 | 验证 Gemini TTS 在 edge_tts 7.x 环境下仍会返回项目兼容的字幕结构， |
| 351 | 并且可以被 `subtitle_provider=edge` 的字幕生成链路直接消费， |
| 352 | 避免再次回退 Whisper。 |
| 353 | """ |
| 354 | |
| 355 | class _InlineData: |
| 356 | def __init__(self, data): |
| 357 | self.data = data |
| 358 | |
| 359 | class _Part: |
| 360 | def __init__(self, data): |
| 361 | self.inline_data = _InlineData(data) |
| 362 | |
| 363 | class _Content: |
| 364 | def __init__(self, data): |
| 365 | self.parts = [_Part(data)] |
| 366 | |
| 367 | class _Candidate: |
| 368 | def __init__(self, data): |
| 369 | self.content = _Content(data) |
| 370 | |
| 371 | class _Response: |
| 372 | def __init__(self, data): |
| 373 | self.candidates = [_Candidate(data)] |
| 374 | |
| 375 | class _FakeModel: |
| 376 | def __init__(self, name): |
| 377 | self.name = name |
| 378 | |
| 379 | def generate_content(self, contents, generation_config): |
| 380 | tone = ( |
| 381 | AudioSegment.silent(duration=1800) |
| 382 | .set_frame_rate(24000) |
| 383 | .set_channels(1) |
| 384 | .set_sample_width(2) |
| 385 | ) |
| 386 | return _Response(tone.raw_data) |
| 387 | |
| 388 | voice_file = f"{temp_dir}/tts-gemini-Zephyr.mp3" |
| 389 | subtitle_file = f"{temp_dir}/tts-gemini-Zephyr.srt" |
| 390 | text = "Gemini subtitle generation should work now. Testing multiple lines." |
| 391 | |
| 392 | with patch("google.generativeai.configure"), patch( |
| 393 | "google.generativeai.GenerativeModel", _FakeModel |
| 394 | ), patch.object(vs.config, "app", dict(vs.config.app, gemini_api_key="test-key")): |
| 395 | sub_maker = vs.gemini_tts( |
| 396 | text=text, |
| 397 | voice_name="Zephyr", |
| 398 | voice_rate=1.0, |
| 399 | voice_file=voice_file, |
| 400 | ) |
| 401 | |
| 402 | self.assertIsNotNone(sub_maker) |
| 403 | self.assertEqual( |
| 404 | getattr(sub_maker, "subs", []), |
| 405 | ["Gemini subtitle generation should work now", "Testing multiple lines"], |
| 406 | ) |
| 407 | self.assertEqual(len(getattr(sub_maker, "offset", [])), 2) |
| 408 | self.assertEqual(sub_maker.offset[0][0], 0) |
| 409 | self.assertLess(sub_maker.offset[0][1], sub_maker.offset[1][1]) |
| 410 | |
| 411 | vs.create_subtitle(sub_maker=sub_maker, text=text, subtitle_file=subtitle_file) |
| 412 | subtitle_content = Path(subtitle_file).read_text(encoding="utf-8") |
| 413 | self.assertIn("Gemini subtitle generation should work now", subtitle_content) |
| 414 | self.assertIn("Testing multiple lines", subtitle_content) |
| 415 | |
| 416 | def test_mimo_tts_uses_openai_compatible_audio_response(self): |
| 417 | """ |
| 418 | 验证 Xiaomi MiMo TTS 可以消费 OpenAI-compatible 的音频响应结构。 |
| 419 | |
| 420 | 这里用 fake OpenAI client 和 fake AudioSegment 覆盖真实网络与 ffmpeg， |
| 421 | 确认运行时代码会把待合成文本放到 assistant message，并把返回的 |
| 422 | base64 WAV 音频导出到项目后续流程使用的音频文件。 |
| 423 | """ |
| 424 | |
| 425 | class _FakeAudio: |
| 426 | def __init__(self): |
| 427 | self.data = base64.b64encode(b"RIFF-fake-wav").decode("utf-8") |
| 428 | |
| 429 | class _FakeMessage: |
| 430 | def __init__(self): |
| 431 | self.audio = _FakeAudio() |
| 432 | |
| 433 | class _FakeChoice: |
| 434 | def __init__(self): |
| 435 | self.message = _FakeMessage() |
| 436 | |
| 437 | class _FakeCompletion: |
| 438 | def __init__(self): |
| 439 | self.choices = [_FakeChoice()] |
| 440 | |
| 441 | class _FakeCompletions: |
| 442 | def create(self, **kwargs): |
| 443 | self.kwargs = kwargs |
| 444 | return _FakeCompletion() |
| 445 | |
| 446 | class _FakeAudioSegment: |
| 447 | def __len__(self): |
| 448 | return 1800 |
| 449 | |
| 450 | def export(self, output_file, format): |
| 451 | Path(output_file).write_bytes(b"fake-mp3") |
| 452 | |
| 453 | fake_completions = _FakeCompletions() |
| 454 | fake_client = SimpleNamespace( |
| 455 | chat=SimpleNamespace(completions=fake_completions) |
| 456 | ) |
| 457 | |
| 458 | with tempfile.TemporaryDirectory() as tmp_dir, patch.object( |
| 459 | vs, |
| 460 | "OpenAI", |
| 461 | return_value=fake_client, |
| 462 | ) as openai_client, patch( |
| 463 | "pydub.AudioSegment.from_file", |
| 464 | return_value=_FakeAudioSegment(), |
| 465 | ), patch.object( |
| 466 | vs.config, |
| 467 | "app", |
| 468 | dict( |
| 469 | vs.config.app, |
| 470 | mimo_api_key="mimo-key", |
| 471 | mimo_base_url="https://api.xiaomimimo.com/v1", |
| 472 | mimo_tts_model_name="mimo-v2.5-tts", |
| 473 | mimo_tts_style_prompt="用清晰的中文旁白朗读。", |
| 474 | ), |
| 475 | ): |
| 476 | voice_file = str(Path(tmp_dir) / "mimo-tts.mp3") |
| 477 | sub_maker = vs.mimo_tts( |
| 478 | text="小米语音合成测试。第二句话。", |
| 479 | voice_name="冰糖", |
| 480 | voice_rate=1.0, |
| 481 | voice_file=voice_file, |
| 482 | voice_volume=1.0, |
| 483 | ) |
| 484 | generated_audio = Path(voice_file).read_bytes() |
| 485 | |
| 486 | openai_client.assert_called_once_with( |
| 487 | api_key="mimo-key", |
| 488 | base_url="https://api.xiaomimimo.com/v1", |
| 489 | ) |
| 490 | self.assertEqual(fake_completions.kwargs["model"], "mimo-v2.5-tts") |
| 491 | self.assertEqual( |
| 492 | fake_completions.kwargs["messages"], |
| 493 | [ |
| 494 | {"role": "user", "content": "用清晰的中文旁白朗读。"}, |
| 495 | {"role": "assistant", "content": "小米语音合成测试。第二句话。"}, |
| 496 | ], |
| 497 | ) |
| 498 | self.assertEqual( |
| 499 | fake_completions.kwargs["audio"], |
| 500 | {"format": "wav", "voice": "冰糖"}, |
| 501 | ) |
| 502 | self.assertEqual(generated_audio, b"fake-mp3") |
| 503 | self.assertIsNotNone(sub_maker) |
| 504 | self.assertEqual(getattr(sub_maker, "subs", []), ["小米语音合成测试", "第二句话"]) |
| 505 | self.assertEqual(len(getattr(sub_maker, "offset", [])), 2) |
| 506 | |
| 507 | def test_chatterbox_voice_helpers(self): |
| 508 | """is_chatterbox_voice / get_chatterbox_voices basics and normalisation.""" |
| 509 | self.assertTrue(vs.is_chatterbox_voice("chatterbox:default-Female")) |
| 510 | self.assertFalse(vs.is_chatterbox_voice("elevenlabs:abc:Rachel")) |
| 511 | self.assertFalse(vs.is_chatterbox_voice("")) |
| 512 | self.assertFalse(vs.is_chatterbox_voice(None)) |
| 513 | |
| 514 | # list entries are normalised to the chatterbox:<name> dispatcher format, |
| 515 | # and entries that are already prefixed are left untouched |
| 516 | with patch.object( |
| 517 | vs.config, |
| 518 | "chatterbox", |
| 519 | {"voices": ["narrator-Male", "chatterbox:host"]}, |
| 520 | ): |
| 521 | self.assertEqual( |
| 522 | vs.get_chatterbox_voices(), |
| 523 | ["chatterbox:narrator-Male", "chatterbox:host"], |
| 524 | ) |
| 525 | |
| 526 | # a comma-separated string is also accepted (TOML-friendly) |
| 527 | with patch.object(vs.config, "chatterbox", {"voices": "alpha, beta ,"}): |
| 528 | self.assertEqual( |
| 529 | vs.get_chatterbox_voices(), |
| 530 | ["chatterbox:alpha", "chatterbox:beta"], |
| 531 | ) |
| 532 | |
| 533 | # with nothing configured the dropdown still gets a usable default |
| 534 | with patch.object(vs.config, "chatterbox", {}): |
| 535 | self.assertEqual(vs.get_chatterbox_voices(), ["chatterbox:default-Female"]) |
| 536 | |
| 537 | def test_chatterbox_tts_posts_to_openai_compatible_endpoint(self): |
| 538 | """Success path: POST /audio/speech, write audio, return legacy SubMaker.""" |
| 539 | |
| 540 | class _FakeResponse: |
| 541 | status_code = 200 |
| 542 | content = b"RIFF-fake-wav" |
| 543 | text = "" |
| 544 | |
| 545 | class _FakeClip: |
| 546 | duration = 3.5 |
| 547 | |
| 548 | def close(self): |
| 549 | pass |
| 550 | |
| 551 | captured = {} |
| 552 | |
| 553 | def _fake_post(url, json=None, headers=None, timeout=None): |
| 554 | captured["url"] = url |
| 555 | captured["json"] = json |
| 556 | captured["headers"] = headers |
| 557 | return _FakeResponse() |
| 558 | |
| 559 | with tempfile.TemporaryDirectory() as tmp_dir, patch.object( |
| 560 | vs.config, |
| 561 | "chatterbox", |
| 562 | { |
| 563 | "base_url": "http://localhost:4123/v1/", |
| 564 | "api_key": "secret", |
| 565 | "model_id": "chatterbox", |
| 566 | }, |
| 567 | ), patch.object( |
| 568 | vs.requests, "post", side_effect=_fake_post |
| 569 | ) as post, patch.object( |
| 570 | vs, "AudioFileClip", return_value=_FakeClip() |
| 571 | ): |
| 572 | voice_file = str(Path(tmp_dir) / "chatterbox.mp3") |
| 573 | sub_maker = vs.chatterbox_tts( |
| 574 | text="Hello world. Second sentence.", |
| 575 | voice="default", |
| 576 | voice_file=voice_file, |
| 577 | voice_rate=1.2, |
| 578 | voice_volume=1.0, |
| 579 | ) |
| 580 | generated_audio = Path(voice_file).read_bytes() |
| 581 | |
| 582 | post.assert_called_once() |
| 583 | # trailing slash on base_url is stripped before appending /audio/speech |
| 584 | self.assertEqual(captured["url"], "http://localhost:4123/v1/audio/speech") |
| 585 | self.assertEqual(captured["json"]["model"], "chatterbox") |
| 586 | self.assertEqual(captured["json"]["voice"], "default") |
| 587 | self.assertEqual(captured["json"]["input"], "Hello world. Second sentence.") |
| 588 | self.assertAlmostEqual(captured["json"]["speed"], 1.2) |
| 589 | # api_key is forwarded as a bearer token |
| 590 | self.assertEqual(captured["headers"].get("Authorization"), "Bearer secret") |
| 591 | # volume is intentionally not part of the OpenAI speech payload |
| 592 | self.assertNotIn("volume", captured["json"]) |
| 593 | self.assertEqual(generated_audio, b"RIFF-fake-wav") |
| 594 | self.assertIsNotNone(sub_maker) |
| 595 | self.assertTrue(getattr(sub_maker, "subs", [])) |
| 596 | |
| 597 | def test_chatterbox_tts_requires_base_url(self): |
| 598 | """Missing base_url short-circuits without any network call.""" |
| 599 | with patch.object( |
| 600 | vs.config, "chatterbox", {"base_url": ""} |
| 601 | ), patch.object(vs.requests, "post") as post: |
| 602 | result = vs.chatterbox_tts( |
| 603 | text="hi", voice="default", voice_file="unused.mp3" |
| 604 | ) |
| 605 | self.assertIsNone(result) |
| 606 | post.assert_not_called() |
| 607 | |
| 608 | def test_chatterbox_tts_returns_none_on_http_error(self): |
| 609 | """A non-200 response is retried up to 3 times, then fails to None.""" |
| 610 | |
| 611 | class _FakeResponse: |
| 612 | status_code = 500 |
| 613 | content = b"" |
| 614 | text = "boom" |
| 615 | |
| 616 | with tempfile.TemporaryDirectory() as tmp_dir, patch.object( |
| 617 | vs.config, "chatterbox", {"base_url": "http://localhost:4123/v1"} |
| 618 | ), patch.object( |
| 619 | vs.requests, "post", return_value=_FakeResponse() |
| 620 | ) as post: |
| 621 | voice_file = str(Path(tmp_dir) / "chatterbox.mp3") |
| 622 | result = vs.chatterbox_tts( |
| 623 | text="hi", voice="default", voice_file=voice_file |
| 624 | ) |
| 625 | self.assertIsNone(result) |
| 626 | self.assertEqual(post.call_count, 3) |
| 627 | |
| 628 | def test_generate_subtitle_keeps_edge_provider_for_gemini_legacy_submaker(self): |
| 629 | """ |
| 630 | 验证 Gemini TTS 返回的 legacy 字幕结构在 edge provider 下可以直接产出 |
| 631 | SRT，不会因为匹配失败而回退到 Whisper。 |
| 632 | """ |
| 633 | script = "Gemini subtitle generation should work now. Testing multiple lines." |
| 634 | sub_maker = vs.populate_legacy_submaker_with_full_text( |
| 635 | vs.ensure_legacy_submaker_fields(vs.SubMaker()), |
| 636 | script, |
| 637 | 2.4, |
| 638 | ) |
| 639 | |
| 640 | with tempfile.TemporaryDirectory() as tmp_dir, patch.object( |
| 641 | task_service.config, |
| 642 | "app", |
| 643 | dict(task_service.config.app, subtitle_provider="edge"), |
| 644 | ), patch("app.services.subtitle.create") as whisper_create, patch( |
| 645 | "app.utils.utils.task_dir", |
| 646 | lambda tid="": str(Path(tmp_dir) / tid) if tid else str(Path(tmp_dir)), |
| 647 | ): |
| 648 | task_id = "gemini-subtitle-edge-task" |
| 649 | Path(tmp_dir, task_id).mkdir(parents=True, exist_ok=True) |
| 650 | subtitle_path = task_service.generate_subtitle( |
| 651 | task_id=task_id, |
| 652 | params=type("Params", (), {"subtitle_enabled": True})(), |
| 653 | video_script=script, |
| 654 | sub_maker=sub_maker, |
| 655 | audio_file="", |
| 656 | ) |
| 657 | |
| 658 | self.assertTrue(subtitle_path.endswith("subtitle.srt")) |
| 659 | self.assertTrue(Path(subtitle_path).exists()) |
| 660 | self.assertFalse(whisper_create.called) |
| 661 | subtitle_content = Path(subtitle_path).read_text(encoding="utf-8") |
| 662 | self.assertIn("Gemini subtitle generation should work now", subtitle_content) |
| 663 | self.assertIn("Testing multiple lines", subtitle_content) |
| 664 | |
| 665 | def test_script_split_keeps_thousand_separator_comma(self): |
| 666 | """ |
| 667 | Edge TTS 会把 "1,000 years" 作为连续文本返回。脚本断句时不能把 |
| 668 | 数字中间的英文逗号当成句子边界，否则字幕聚合会出现 issue #894 |
| 669 | 里的 sub_items 数量少于 script_lines，并错误回退 Whisper。 |
| 670 | """ |
| 671 | text = ( |
| 672 | "It takes about 1,000 years for a single drop of water to finish " |
| 673 | "the whole trip!" |
| 674 | ) |
| 675 | |
| 676 | self.assertEqual( |
| 677 | utils.split_string_by_punctuations(text), |
| 678 | [ |
| 679 | ( |
| 680 | "It takes about 1,000 years for a single drop of water to finish " |
| 681 | "the whole trip" |
| 682 | ) |
| 683 | ], |
| 684 | ) |
| 685 | |
| 686 | def test_edge_cue_aggregation_handles_thousand_separator_comma(self): |
| 687 | """ |
| 688 | 复现 issue #894 的关键形态：Edge cues 中最后一句作为连续文本返回， |
| 689 | 包含 `1,000 years`。脚本断句必须与 cues 聚合结果一致，不能把它 |
| 690 | 拆成两条字幕。 |
| 691 | """ |
| 692 | text = ( |
| 693 | "The ocean isn't just sitting stil, it moves around the world like a massive " |
| 694 | "amusement park ride! Cold water at the North and South Poles sinks to the " |
| 695 | "bottom because it is heavy and salty. At the same time, warm water from the " |
| 696 | "sunny equator flows along the top to take its place. This creates a giant " |
| 697 | "underwater conveyor belt that travels all the way around the Earth. It takes " |
| 698 | "about 1,000 years for a single drop of water to finish the whole trip!" |
| 699 | ) |
| 700 | script_lines = utils.split_string_by_punctuations(text) |
| 701 | cues = [] |
| 702 | for index, line in enumerate(script_lines): |
| 703 | # Edge 的 cue content 经常没有脚本里的空格和标点布局，这里去掉空格 |
| 704 | # 来模拟更严格的匹配场景。 |
| 705 | cues.append( |
| 706 | SimpleNamespace( |
| 707 | content=line.replace(" ", ""), |
| 708 | start=timedelta(seconds=index), |
| 709 | end=timedelta(seconds=index + 0.8), |
| 710 | ) |
| 711 | ) |
| 712 | sub_maker = SimpleNamespace(cues=cues) |
| 713 | |
| 714 | sub_items = vs._build_subtitle_items_from_edge_cues(sub_maker, script_lines) |
| 715 | |
| 716 | self.assertEqual(len(sub_items), len(script_lines)) |
| 717 | self.assertIn("1,000 years", sub_items[-1]) |
| 718 | |
| 719 | def test_script_split_supports_arabic_punctuation(self): |
| 720 | """ |
| 721 | 阿拉伯语脚本常用 ، ؛ ؟ 作为自然断句标点。断句阶段必须识别这些 |
| 722 | 标点，否则 edge-tts cue 的停顿边界和脚本行边界会错位。 |
| 723 | """ |
| 724 | text = "مرحبا بالعالم، كيف حالك؟ هذا اختبار؛ يعمل بشكل جيد." |
| 725 | |
| 726 | self.assertEqual( |
| 727 | utils.split_string_by_punctuations(text), |
| 728 | [ |
| 729 | "مرحبا بالعالم", |
| 730 | "كيف حالك", |
| 731 | "هذا اختبار", |
| 732 | "يعمل بشكل جيد", |
| 733 | ], |
| 734 | ) |
| 735 | |
| 736 | def test_match_script_line_normalizes_arabic_letter_forms(self): |
| 737 | """ |
| 738 | edge-tts 可能把阿拉伯语中的不同字母形态归一化，或返回带变音符号、 |
| 739 | Tatweel 的 cue 文本。匹配时应容错，但最终字幕仍保留原始脚本文案。 |
| 740 | """ |
| 741 | script_lines = ["أهلاً وسهلاً بك في المدرسة"] |
| 742 | |
| 743 | matched = vs._match_script_line( |
| 744 | script_lines, |
| 745 | "اهلا وسهلا بك في المدرسه", |
| 746 | 0, |
| 747 | ) |
| 748 | |
| 749 | self.assertEqual(matched, script_lines[0]) |
| 750 | |
| 751 | def test_edge_cue_aggregation_handles_arabic_variant_forms(self): |
| 752 | """ |
| 753 | 复现阿拉伯语字幕失败的核心路径：脚本包含 أ/ة 等字母形态，edge cue |
| 754 | 返回 ا/ه 等归一化形态时，聚合仍应生成完整字幕，避免回退 Whisper。 |
| 755 | """ |
| 756 | text = "أهلاً وسهلاً بك في المدرسة؟ هذا اختبار رائع، شكراً لك." |
| 757 | script_lines = utils.split_string_by_punctuations(text) |
| 758 | cue_texts = [ |
| 759 | "اهلا وسهلا بك في المدرسه", |
| 760 | "هذا اختبار رائع", |
| 761 | "شكرا لك", |
| 762 | ] |
| 763 | sub_maker = SimpleNamespace( |
| 764 | cues=[ |
| 765 | SimpleNamespace( |
| 766 | content=cue_text, |
| 767 | start=timedelta(seconds=index), |
| 768 | end=timedelta(seconds=index + 0.8), |
| 769 | ) |
| 770 | for index, cue_text in enumerate(cue_texts) |
| 771 | ] |
| 772 | ) |
| 773 | |
| 774 | sub_items = vs._build_subtitle_items_from_edge_cues(sub_maker, script_lines) |
| 775 | |
| 776 | self.assertEqual(len(sub_items), len(script_lines)) |
| 777 | self.assertIn("أهلاً وسهلاً بك في المدرسة", sub_items[0]) |
| 778 | self.assertIn("شكراً لك", sub_items[-1]) |
| 779 | |
| 780 | def test_create_subtitle_ignores_markdown_separator_lines(self): |
| 781 | """ |
| 782 | 用户手动脚本可能包含 `---` 这类 Markdown 分隔符。TTS 不会朗读 |
| 783 | 这些符号行，字幕聚合也不应把它们当成目标字幕行，否则后续真实 |
| 784 | 字幕会卡住并回退到 Whisper。 |
| 785 | """ |
| 786 | text = "第一段\n---\n第二段" |
| 787 | sub_maker = SimpleNamespace( |
| 788 | cues=[ |
| 789 | SimpleNamespace( |
| 790 | content="第一段", |
| 791 | start=timedelta(seconds=0), |
| 792 | end=timedelta(seconds=0.8), |
| 793 | ), |
| 794 | SimpleNamespace( |
| 795 | content="第二段", |
| 796 | start=timedelta(seconds=1), |
| 797 | end=timedelta(seconds=1.8), |
| 798 | ), |
| 799 | ] |
| 800 | ) |
| 801 | |
| 802 | with tempfile.TemporaryDirectory() as tmp_dir: |
| 803 | subtitle_file = Path(tmp_dir) / "subtitle.srt" |
| 804 | vs.create_subtitle( |
| 805 | sub_maker=sub_maker, |
| 806 | text=text, |
| 807 | subtitle_file=str(subtitle_file), |
| 808 | ) |
| 809 | |
| 810 | subtitle_content = subtitle_file.read_text(encoding="utf-8") |
| 811 | |
| 812 | self.assertIn("第一段", subtitle_content) |
| 813 | self.assertIn("第二段", subtitle_content) |
| 814 | self.assertNotIn("---", subtitle_content) |
| 815 | self.assertNotIn("00:00:00,000 --> 00:00:00,000", subtitle_content) |
| 816 | |
| 817 | def test_create_subtitle_ignores_markdown_underscore_marks(self): |
| 818 | """ |
| 819 | `_` 常被用户用作 Markdown 强调标记，但 TTS 返回的 cue 通常不包含 |
| 820 | 这些格式符。匹配时应忽略 `_`，避免生成空字幕或回退到 Whisper。 |
| 821 | """ |
| 822 | text = "这是_a_测试。" |
| 823 | sub_maker = SimpleNamespace( |
| 824 | cues=[ |
| 825 | SimpleNamespace( |
| 826 | content="这是a测试", |
| 827 | start=timedelta(seconds=0), |
| 828 | end=timedelta(seconds=0.8), |
| 829 | ), |
| 830 | ] |
| 831 | ) |
| 832 | |
| 833 | with tempfile.TemporaryDirectory() as tmp_dir: |
| 834 | subtitle_file = Path(tmp_dir) / "subtitle.srt" |
| 835 | vs.create_subtitle( |
| 836 | sub_maker=sub_maker, |
| 837 | text=text, |
| 838 | subtitle_file=str(subtitle_file), |
| 839 | ) |
| 840 | |
| 841 | subtitle_content = subtitle_file.read_text(encoding="utf-8") |
| 842 | |
| 843 | self.assertIn("这是a测试", subtitle_content) |
| 844 | self.assertNotIn("这是_a_测试", subtitle_content) |
| 845 | self.assertNotIn("00:00:00,000 --> 00:00:00,000", subtitle_content) |
| 846 | |
| 847 | def test_convert_rate_to_percent_signs_zero_rate(self): |
| 848 | # Rates near but not exactly 1.0 round to 0 percent. edge-tts rejects |
| 849 | # an unsigned "0%" (ValueError: Invalid rate '0%'), so the helper must |
| 850 | # emit a sign-prefixed "+0%". Regression test for that crash. |
| 851 | self.assertEqual(vs.convert_rate_to_percent(1.0), "+0%") |
| 852 | self.assertEqual(vs.convert_rate_to_percent(1.004), "+0%") |
| 853 | self.assertEqual(vs.convert_rate_to_percent(0.997), "+0%") |
| 854 | self.assertEqual(vs.convert_rate_to_percent(1.5), "+50%") |
| 855 | self.assertEqual(vs.convert_rate_to_percent(0.8), "-20%") |
| 856 | |
| 857 | def test_convert_rate_to_percent_invalid_values_default_to_normal(self): |
| 858 | # API 和批处理脚本可能把空语速传成 0、None 或空字符串；这些都不应让 |
| 859 | # edge-tts 收到 -100% 或触发异常，而是按正常语速处理。 |
| 860 | self.assertEqual(vs.convert_rate_to_percent(0), "+0%") |
| 861 | self.assertEqual(vs.convert_rate_to_percent(0.0), "+0%") |
| 862 | self.assertEqual(vs.convert_rate_to_percent(None), "+0%") |
| 863 | self.assertEqual(vs.convert_rate_to_percent(""), "+0%") |
| 864 | |
| 865 | |
| 866 | class TestElevenLabsVoice(unittest.TestCase): |
| 867 | |
| 868 | def test_is_elevenlabs_voice_true(self): |
| 869 | self.assertTrue(vs.is_elevenlabs_voice("elevenlabs:pNInz6obpgDQGcFmaJgB:Adam")) |
| 870 | |
| 871 | def test_is_elevenlabs_voice_false_azure(self): |
| 872 | self.assertFalse(vs.is_elevenlabs_voice("zh-CN-XiaoxiaoNeural-Female")) |
| 873 | |
| 874 | def test_is_elevenlabs_voice_false_siliconflow(self): |
| 875 | self.assertFalse(vs.is_elevenlabs_voice("siliconflow:model:voice-Male")) |
| 876 | |
| 877 | def test_is_elevenlabs_voice_empty(self): |
| 878 | self.assertFalse(vs.is_elevenlabs_voice("")) |
| 879 | |
| 880 | def test_is_elevenlabs_voice_none(self): |
| 881 | self.assertFalse(vs.is_elevenlabs_voice(None)) |
| 882 | |
| 883 | def test_get_elevenlabs_voices_empty_api_key(self): |
| 884 | result = vs.get_elevenlabs_voices("") |
| 885 | self.assertEqual(result, []) |
| 886 | |
| 887 | @patch("app.services.voice.requests.get") |
| 888 | def test_get_elevenlabs_voices_success(self, mock_get): |
| 889 | mock_get.return_value.status_code = 200 |
| 890 | mock_get.return_value.json.return_value = { |
| 891 | "voices": [ |
| 892 | {"voice_id": "abc123", "name": "Adam"}, |
| 893 | {"voice_id": "def456", "name": "Rachel"}, |
| 894 | ] |
| 895 | } |
| 896 | result = vs.get_elevenlabs_voices("fake-api-key") |
| 897 | self.assertEqual(result, [ |
| 898 | "elevenlabs:abc123:Adam", |
| 899 | "elevenlabs:def456:Rachel", |
| 900 | ]) |
| 901 | mock_get.assert_called_once() |
| 902 | call_kwargs = mock_get.call_args |
| 903 | self.assertIn("xi-api-key", call_kwargs.kwargs.get("headers", {})) |
| 904 | |
| 905 | @patch("app.services.voice.requests.get") |
| 906 | def test_get_elevenlabs_voices_http_error(self, mock_get): |
| 907 | mock_get.return_value.status_code = 401 |
| 908 | mock_get.return_value.text = "Unauthorized" |
| 909 | result = vs.get_elevenlabs_voices("bad-key") |
| 910 | self.assertEqual(result, []) |
| 911 | |
| 912 | @patch("app.services.voice.requests.get") |
| 913 | def test_get_elevenlabs_voices_network_error(self, mock_get): |
| 914 | import requests as req_lib |
| 915 | mock_get.side_effect = req_lib.exceptions.ConnectionError("timeout") |
| 916 | result = vs.get_elevenlabs_voices("fake-key") |
| 917 | self.assertEqual(result, []) |
| 918 | |
| 919 | @patch("app.services.voice.requests.post") |
| 920 | @patch("app.services.voice.AudioFileClip") |
| 921 | @patch("app.services.voice.config") |
| 922 | def test_elevenlabs_tts_success(self, mock_config, mock_clip_cls, mock_post): |
| 923 | mock_config.elevenlabs.get.return_value = "fake-api-key" |
| 924 | mock_post.return_value.status_code = 200 |
| 925 | mock_post.return_value.content = b"fake-mp3-bytes" |
| 926 | mock_clip = mock_clip_cls.return_value.__enter__.return_value |
| 927 | mock_clip_cls.return_value.duration = 3.0 |
| 928 | mock_clip_cls.return_value.close = lambda: None |
| 929 | |
| 930 | with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as f: |
| 931 | out_path = f.name |
| 932 | |
| 933 | try: |
| 934 | result = vs.elevenlabs_tts("Hello world", "abc123", out_path) |
| 935 | self.assertIsNotNone(result) |
| 936 | self.assertTrue(hasattr(result, "subs")) |
| 937 | self.assertTrue(hasattr(result, "offset")) |
| 938 | finally: |
| 939 | if os.path.exists(out_path): |
| 940 | os.remove(out_path) |
| 941 | |
| 942 | @patch("app.services.voice.config") |
| 943 | def test_elevenlabs_tts_no_api_key(self, mock_config): |
| 944 | mock_config.elevenlabs.get.return_value = "" |
| 945 | result = vs.elevenlabs_tts("Hello", "abc123", "/tmp/test.mp3") |
| 946 | self.assertIsNone(result) |
| 947 | |
| 948 | @patch("app.services.voice.config") |
| 949 | def test_elevenlabs_tts_empty_text(self, mock_config): |
| 950 | mock_config.elevenlabs.get.return_value = "fake-key" |
| 951 | result = vs.elevenlabs_tts(" ", "abc123", "/tmp/test.mp3") |
| 952 | self.assertIsNone(result) |
| 953 | |
| 954 | |
| 955 | if __name__ == "__main__": |
| 956 | # python -m unittest test.services.test_voice.TestVoiceService.test_azure_tts_v1 |
| 957 | # python -m unittest test.services.test_voice.TestVoiceService.test_azure_tts_v2 |
| 958 | unittest.main() |
| 959 |