feat: transcribe_voice 新增 OpenAI Whisper API 后端 (#66)
默认 local，零行为变化。opt-in 双因素：transcription_backend=openai 且 openai_api_key 都齐才生效；任一缺失静默回退 local + stderr 一行警告。 首次进入云路径会 stderr 警告"语音将上传至 OpenAI 服务器"。 新增 config.json 字段： - transcription_backend: "local" (默认) | "openai" - local_whisper_model: "base" (替换 mcp_server.py 里硬编码 DEFAULT_WHISPER_MODEL) - openai_api_key: "" (默认空；openai 包为 optional，按需 pip install) 关键技术选择： - _transcribe(wav, backend) 单一 if/else 分发，不引入插件/工厂层 (Rule of Three —— 只有一个云后端时不值得抽象) - 文件 > 25MB 在 OpenAI() 实例化之前提前拒绝，避免无谓上传 - 错误分类清晰: 缺 key / 缺 openai 包 / 401 / 429 / APIError 各自的提示 - PR #58 缓存 schema 自然扩展: 条目加 backend 字段，命中需 backend+model_size 都匹配 - 旧条目缺 backend 字段视为 "local"，向前兼容 PR #58 已落盘的所有数据 - transcribe_chat.py 批量 CLI 与 MCP 工具共享同一份配置，保持一致 新增 2 个测试 (tests/test_openai_backend.py)，只覆盖回归风险最高的两条： - 文件 > 25MB 必须在 SDK 实例化前拒绝（隐私契约的防线） - backend 不匹配的旧条目不命中（避免切后端时返回错后端结果） 其余路径要么琐碎（默认值读取）、要么坏掉时声音很大（SDK 错误、ImportError）， 要么已被 PR #58 现有测试隐式覆盖（缺 backend 字段的旧条目），不再单独写测试。 顺手把 README 里 PR #53 漏掉的 voice 三件套（get_voice_messages / decode_voice / transcribe_voice）补进 MCP 工具表，并新增"⚠️ 语音转录隐私" 章节说清数据流向、成本（约 \$0.006/分钟）、25MB 上限、回退行为。 Closes ylytdeng/wechat-decrypt#59
This commit is contained in:
btc-z
116
tests/test_openai_backend.py
Normal file
116
tests/test_openai_backend.py
Normal file
@@ -0,0 +1,116 @@
|
||||
"""
|
||||
issue #59: opt-in OpenAI Whisper API 后端的两条关键回归测试。
|
||||
|
||||
只测两件事：
|
||||
1. 隐私契约: 文件 > 25MB 在调用 OpenAI SDK 之前就被拒绝（保证不会无意上传）
|
||||
2. 缓存正确性: backend 不匹配的旧条目不会被命中（避免切后端时返回错后端结果）
|
||||
|
||||
其余路径要么琐碎（默认值读取）、要么坏掉时声音很大（SDK 错误、ImportError），
|
||||
不再单独覆盖。
|
||||
"""
|
||||
import os
|
||||
import sys
|
||||
import tempfile
|
||||
import unittest
|
||||
from unittest.mock import MagicMock, patch
|
||||
|
||||
import mcp_server
|
||||
|
||||
|
||||
class _CacheIsolationMixin:
|
||||
"""与 test_voice_transcription_cache.py 同款隔离：避免污染 module-level 缓存状态。"""
|
||||
|
||||
def setUp(self):
|
||||
self._saved_cache = mcp_server._voice_transcription_cache
|
||||
self._saved_path = mcp_server.VOICE_TRANSCRIPTION_CACHE_FILE
|
||||
self._saved_warned = mcp_server._voice_transcription_save_warned
|
||||
|
||||
mcp_server._voice_transcription_cache = None
|
||||
mcp_server._voice_transcription_save_warned = False
|
||||
|
||||
self._tmp = tempfile.TemporaryDirectory()
|
||||
self.addCleanup(self._tmp.cleanup)
|
||||
mcp_server.VOICE_TRANSCRIPTION_CACHE_FILE = os.path.join(
|
||||
self._tmp.name, "voice_transcriptions.json"
|
||||
)
|
||||
|
||||
def tearDown(self):
|
||||
mcp_server._voice_transcription_cache = self._saved_cache
|
||||
mcp_server.VOICE_TRANSCRIPTION_CACHE_FILE = self._saved_path
|
||||
mcp_server._voice_transcription_save_warned = self._saved_warned
|
||||
|
||||
|
||||
class OpenAIBackendPrivacyTests(unittest.TestCase):
|
||||
"""隐私契约：超限文件必须在 OpenAI SDK 实例化之前就被拒绝。
|
||||
|
||||
若有人把 size check 移到 OpenAI(api_key=...) 之后（即便仍在 upload 前），
|
||||
本测试会失败 —— 这层防御边界值得守住。
|
||||
"""
|
||||
|
||||
def test_oversize_audio_rejected_before_sdk_call(self):
|
||||
# 写一个 26MB 临时 WAV (用稀疏写法快速生成)
|
||||
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
|
||||
f.seek(26 * 1024 * 1024)
|
||||
f.write(b"\0")
|
||||
big_path = f.name
|
||||
self.addCleanup(os.unlink, big_path)
|
||||
|
||||
# 注入一个假的 openai 模块，保证 import 成功；OpenAI 构造函数若被调用即测试失败
|
||||
fake_openai = MagicMock()
|
||||
fake_openai.OpenAI = MagicMock(
|
||||
side_effect=AssertionError("OpenAI() must not be instantiated for oversize files")
|
||||
)
|
||||
fake_openai.AuthenticationError = type("AuthenticationError", (Exception,), {})
|
||||
fake_openai.RateLimitError = type("RateLimitError", (Exception,), {})
|
||||
fake_openai.APIError = type("APIError", (Exception,), {})
|
||||
|
||||
with patch.dict(sys.modules, {"openai": fake_openai}):
|
||||
with self.assertRaises(RuntimeError) as ctx:
|
||||
mcp_server._transcribe_openai(big_path)
|
||||
|
||||
self.assertIn("25MB", str(ctx.exception))
|
||||
fake_openai.OpenAI.assert_not_called()
|
||||
|
||||
|
||||
class CacheBackendMatchTests(_CacheIsolationMixin, unittest.TestCase):
|
||||
"""缓存正确性：backend 不匹配 → 视为 miss，避免切后端时返回错后端结果。"""
|
||||
|
||||
def test_cache_hit_requires_backend_match(self):
|
||||
# 种入一条 openai 后端的缓存条目
|
||||
key = mcp_server._voice_transcription_cache_key("wxid_test", 42)
|
||||
cache = mcp_server._load_voice_transcription_cache()
|
||||
cache[key] = {
|
||||
"text": "openai-result",
|
||||
"language": "zh",
|
||||
"create_time": 1700000000,
|
||||
"backend": "openai",
|
||||
"model_size": "whisper-1",
|
||||
}
|
||||
mcp_server._save_voice_transcription_cache()
|
||||
|
||||
# 当前后端是 local，应当 miss → 走转录流程而非返回 "openai-result"
|
||||
with patch.object(mcp_server, "TRANSCRIPTION_BACKEND", "local"), \
|
||||
patch.object(mcp_server, "OPENAI_API_KEY", ""), \
|
||||
patch.object(mcp_server, "resolve_username", return_value="wxid_test"), \
|
||||
patch.object(mcp_server, "_fetch_voice_row",
|
||||
return_value=(b"\x02fake-silk-blob", 1700000001)), \
|
||||
patch.object(mcp_server, "_silk_to_wav",
|
||||
return_value=("/tmp/fake.wav", 24000 * 2)), \
|
||||
patch.object(mcp_server, "_transcribe_local",
|
||||
return_value={"text": "local-result", "language": "zh"}), \
|
||||
patch.dict(sys.modules, {"whisper": MagicMock(), "pysilk": MagicMock()}):
|
||||
result = mcp_server.transcribe_voice("test_contact", 42)
|
||||
|
||||
# 没返回旧 openai 缓存，而是走了 local 转录流程
|
||||
self.assertNotIn("openai-result", result)
|
||||
self.assertIn("local-result", result)
|
||||
|
||||
# 落盘的新条目应记录当前后端
|
||||
mcp_server._voice_transcription_cache = None
|
||||
reloaded = mcp_server._load_voice_transcription_cache()
|
||||
self.assertEqual(reloaded[key]["backend"], "local")
|
||||
self.assertEqual(reloaded[key]["text"], "local-result")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -194,7 +194,7 @@ class TranscribeVoiceCacheHitTests(_CacheIsolationMixin, unittest.TestCase):
|
||||
"text": "缓存命中文本",
|
||||
"language": "zh",
|
||||
"create_time": 1700000000,
|
||||
"model_size": mcp_server.DEFAULT_WHISPER_MODEL,
|
||||
"model_size": mcp_server.LOCAL_WHISPER_MODEL,
|
||||
})
|
||||
|
||||
with patch.object(mcp_server, "resolve_username", return_value="wxid_test") as mock_resolve, \
|
||||
@@ -216,7 +216,7 @@ class TranscribeVoiceCacheHitTests(_CacheIsolationMixin, unittest.TestCase):
|
||||
self._seed(key, {
|
||||
"text": "历史条目",
|
||||
"language": "zh",
|
||||
"model_size": mcp_server.DEFAULT_WHISPER_MODEL,
|
||||
"model_size": mcp_server.LOCAL_WHISPER_MODEL,
|
||||
})
|
||||
|
||||
with patch.object(mcp_server, "resolve_username", return_value="wxid_test"), \
|
||||
@@ -233,7 +233,7 @@ class TranscribeVoiceCacheHitTests(_CacheIsolationMixin, unittest.TestCase):
|
||||
"text": "",
|
||||
"language": "zh",
|
||||
"create_time": 1700000000,
|
||||
"model_size": mcp_server.DEFAULT_WHISPER_MODEL,
|
||||
"model_size": mcp_server.LOCAL_WHISPER_MODEL,
|
||||
})
|
||||
|
||||
with patch.object(mcp_server, "resolve_username", return_value="wxid_test"), \
|
||||
@@ -244,7 +244,7 @@ class TranscribeVoiceCacheHitTests(_CacheIsolationMixin, unittest.TestCase):
|
||||
self.assertIn("(zh)", result)
|
||||
|
||||
def test_model_mismatch_is_treated_as_miss(self):
|
||||
# 缓存条目的 model_size 和当前 DEFAULT_WHISPER_MODEL 不一致时，
|
||||
# 缓存条目的 model_size 和当前 LOCAL_WHISPER_MODEL 不一致时，
|
||||
# 不应命中；进入 miss 路径（这里无 whisper 依赖，应落到"缺少依赖"分支）。
|
||||
key = mcp_server._voice_transcription_cache_key("wxid_test", 10)
|
||||
self._seed(key, {
|
||||
|
||||
Reference in New Issue
Block a user