Files
xieyi-weixin-/test_protocol_voice.py
T
2026-09-21 10:56:48 +08:00

252 lines
12 KiB
Python

"""Exercise transcription -> question/risk/model/send using only synthetic transports."""
import time
import unittest
from unittest import mock
from protocol_engine import session_key, model_context_text
import test_native_protocol_pipeline as helpers
from test_native_protocol_pipeline import message, ACCOUNT, CONV
class ProtocolVoiceTests(unittest.TestCase):
setUp = helpers.EngineTests.setUp
make_bot = helpers.EngineTests.make_bot
enqueue = helpers.EngineTests.enqueue
def voice(self, n=2, status='pending', text='语音内容合成测试', **kwargs):
m = message(n, **kwargs)
m.update(content_type=16, voice_status=status, content='[语音待转文字]')
if status == 'ready':
m.update(voice_transcribed=True, content='[语音转文字] ' + text,
voice_binding=f"{m['account']}\0{m['conv_id']}\0{m['server_id']}")
return m
def test_transcribed_voice_reaches_model_and_selective_review_as_text(self):
from protocol_engine import ProtocolBot
self.bot._needs_review = lambda state: ProtocolBot._needs_review(self.bot, state)
self.bot._assistant_pending = lambda state: False
v = self.voice(status='ready')
with mock.patch('review_policy.selective_review_reason', return_value='合成审核规则') as policy:
self.enqueue(v)
self.gen.assert_called_once()
context = self.gen.call_args.args[0]
self.assertIn('语音内容合成测试', model_context_text(context))
state = self.bot._pending[session_key(ACCOUNT, CONV)]
self.assertIn('语音内容合成测试', policy.call_args.args[1])
self.assertTrue(state['awaiting_review'])
self.assertFalse(state.get('manual_required'))
self.assertFalse(self.sender.calls)
def test_pending_voice_delays_only_its_conversation_then_resumes_on_cache_update(self):
v = self.voice()
self.enqueue(v)
key = session_key(ACCOUNT, CONV)
self.assertEqual(self.bot._pending[key]['stage'], 'voice_transcribing')
self.gen.assert_not_called()
self.db.add(message(4, conv='S:100_201'))
self.bot.poll_once()
self.assertEqual([c[2] for c in self.sender.calls], ['S:100_201'])
v.update(self.voice(status='ready')) # Same server ID/row; no new-message event.
self.bot._pending[key]['ready_at'] = 0
self.bot.poll_once()
self.assertEqual([c[2] for c in self.sender.calls], ['S:100_201', CONV])
self.assertFalse(self.bot.stop_event.is_set())
def test_mixed_voice_and_text_waits_and_preserves_both(self):
v = self.voice()
self.db.add(v)
self.enqueue(message(3, text='补充文字'))
key = session_key(ACCOUNT, CONV)
self.assertEqual(self.bot._pending[key]['stage'], 'voice_transcribing')
v.update(self.voice(status='ready'))
self.bot._pending[key]['ready_at'] = 0
self.bot.poll_once()
context = self.gen.call_args.args[0]
self.assertIn('补充文字', model_context_text(context))
self.assertIn('语音内容合成测试', model_context_text(context))
self.assertEqual(len(self.sender.calls), 1)
def test_failed_transcription_enters_manual_and_never_calls_model(self):
v = self.voice(status='error')
v['voice_error_code'] = 'empty_audio'
self.enqueue(v)
state = self.bot._pending[session_key(ACCOUNT, CONV)]
self.assertTrue(state['manual_required'])
self.assertIn('empty_audio', state['manual_reason'])
self.gen.assert_not_called()
self.assertFalse(self.sender.calls)
def test_missing_voice_times_out_without_global_stop_or_guess(self):
self.enqueue(self.voice(status='missing'))
state = self.bot._pending[session_key(ACCOUNT, CONV)]
state.update(voice_wait_started_at=time.time()-181, ready_at=0)
self.bot.poll_once()
self.assertTrue(state['manual_required'])
self.assertIn('超时', state['manual_reason'])
self.assertFalse(self.bot.stop_event.is_set())
self.gen.assert_not_called()
def test_human_reply_while_voice_pending_discards_old_question(self):
self.enqueue(self.voice())
self.db.add(message(3, self=True, text='人工答复'))
self.bot._pending[session_key(ACCOUNT, CONV)]['ready_at'] = 0
self.bot.poll_once()
self.assertFalse(self.bot._pending)
self.gen.assert_not_called()
self.assertFalse(self.sender.calls)
def test_new_media_after_voice_still_needs_manual(self):
self.db.add(self.voice(status='ready'))
image = message(3)
image['content_type'] = 3
self.enqueue(image)
self.assertTrue(self.bot._pending[session_key(ACCOUNT, CONV)]['manual_required'])
self.gen.assert_not_called()
def test_spoofed_or_cross_conversation_transcription_does_not_reach_model(self):
v = self.voice(status='ready')
v['voice_binding'] = '101\0M:200\0'+'2'
self.enqueue(v)
self.gen.assert_not_called()
self.assertFalse(self.sender.calls)
self.assertEqual(self.bot._pending[session_key(ACCOUNT, CONV)]['stage'], 'voice_transcribing')
def test_direct_judge_receives_entire_mixed_unanswered_batch(self):
from protocol_engine import ProtocolBot
self.bot.generate=self.bot._generate
self.bot._needs_review=lambda state:ProtocolBot._needs_review(self.bot,state)
self.bot._assistant_pending=lambda state:False
v=self.voice(status='ready',text='危险语音合成内容,需要人工判断')
self.db.add(v)
result={'reply':'合成待审草稿','judge':{'risk':'high'},'judge_mode':'score_only'}
with mock.patch('ai_config.AI_ENABLED',True), \
mock.patch('ai_config.AI_JUDGE_ENABLED',True), \
mock.patch('ai_config.AI_JUDGE_MODE','score_only'), \
mock.patch('ai_config.AI_REVIEW_RULES',[]), \
mock.patch('ai_chat.gateway_provider',return_value=None), \
mock.patch('ai_chat.current_provider',return_value=object()), \
mock.patch('model_router.answer',return_value=result) as answer:
self.enqueue(message(3,text='好的'))
state=self.bot._pending[session_key(ACCOUNT,CONV)]
customer_text=answer.call_args.kwargs['customer_text']
self.assertEqual(customer_text,state['staged_user_text'])
self.assertIn('危险语音合成内容,需要人工判断',customer_text)
self.assertTrue(customer_text.endswith('好的'))
self.assertTrue(state['awaiting_review'])
self.assertFalse(self.sender.calls)
def test_same_voice_cache_changes_during_generation_discard_then_regenerate(self):
v=self.voice(status='ready',text='首次转写')
generated=[]
def generate(context):
generated.append(model_context_text(context))
if len(generated)==1:
v['content']='[语音转文字] 修正后的转写'
return '旧答复'
return '新答复'
self.bot.generate=generate
self.enqueue(v)
state=self.bot._pending[session_key(ACCOUNT,CONV)]
self.assertEqual(state['stage'],'queued')
self.assertFalse(state['reply_text'])
self.assertFalse(state.get('approved'))
self.assertFalse(self.sender.calls)
self.bot.poll_once()
self.assertEqual(len(generated),2)
self.assertIn('修正后的转写',generated[1])
self.assertEqual([call[3] for call in self.sender.calls],['新答复'])
def test_final_native_guard_rejects_changed_voice_before_send(self):
v=self.voice(status='ready',text='首次转写')
original=self.sender.send
changed=[]
def send(*args,**kwargs):
if not changed:
changed.append(True)
v['content']='[语音转文字] 最后校验前更新'
return original(*args,**kwargs)
self.sender.send=send
self.gen.side_effect=['旧答复','新答复']
self.enqueue(v)
state=self.bot._pending[session_key(ACCOUNT,CONV)]
self.assertEqual(state['stage'],'queued')
self.assertFalse(state['send_state'])
self.assertFalse(state['reply_text'])
self.assertFalse(state.get('approved'))
self.assertFalse(self.sender.calls)
self.bot.poll_once()
self.assertEqual(self.gen.call_count,2)
self.assertEqual([call[3] for call in self.sender.calls],['新答复'])
def test_changed_voice_revokes_approval_and_rechecks_review_on_new_answer(self):
self.bot.send_mode='review'
v=self.voice(status='ready',text='首次转写')
self.enqueue(v)
key=session_key(ACCOUNT,CONV)
self.bot.approve_pending_replies([key])
v['content']='[语音转文字] 审核后内容改变'
self.bot.poll_once()
state=self.bot._pending[key]
self.assertFalse(state.get('approved'))
self.assertFalse(state['reply_text'])
self.assertFalse(self.sender.calls)
self.bot.poll_once()
self.assertTrue(state['awaiting_review'])
self.assertFalse(state.get('approved'))
self.assertEqual(self.gen.call_count,2)
self.assertFalse(self.sender.calls)
self.bot.approve_pending_replies([key])
self.bot.poll_once()
self.assertEqual(len(self.sender.calls),1)
def test_cache_change_keeps_independent_manual_block(self):
v=self.voice(status='ready')
key=session_key(ACCOUNT,CONV)
def generate(context):
v['content']='[语音转文字] 变化后的内容'
self.bot._pending[key].update(manual_required=True,manual_reason='独立人工条件',stage='error')
return '应丢弃的答复'
self.bot.generate=mock.Mock(side_effect=generate)
self.enqueue(v)
state=self.bot._pending[key]
self.assertEqual(state['stage'],'error')
self.assertTrue(state['manual_required'])
self.assertEqual(state['manual_reason'],'独立人工条件')
self.assertFalse(state['reply_text'])
self.bot.poll_once()
self.bot.generate.assert_called_once()
self.assertFalse(self.sender.calls)
def test_unknown_delivery_keeps_original_voice_draft_and_evidence(self):
from wecom_native_sender import DeliveryUnknown
v=self.voice(status='ready',text='首次转写')
self.sender.error=DeliveryUnknown('合成未知回执')
self.enqueue(v)
state=self.bot._pending[session_key(ACCOUNT,CONV)]
original=dict(state)
v['content']='[语音转文字] 之后缓存变化'
self.bot.poll_once()
self.assertEqual(state['send_state'],'uncertain')
self.assertEqual(state['reply_text'],original['reply_text'])
self.assertEqual(state['voice_batch_fingerprint'],original['voice_batch_fingerprint'])
self.assertEqual(len(self.sender.calls),1)
def test_legacy_voice_draft_without_snapshot_regenerates_before_send(self):
self.bot.send_delay_seconds=600
self.enqueue(self.voice(status='ready'))
state=self.bot._pending[session_key(ACCOUNT,CONV)]
state.pop('voice_batch_fingerprint')
state['send_ready_at']=0
self.bot.poll_once()
self.assertFalse(state['reply_text'])
self.assertFalse(self.sender.calls)
self.bot.send_delay_seconds=0
self.bot.poll_once()
self.assertEqual(self.gen.call_count,2)
self.assertEqual(len(self.sender.calls),1)
if __name__ == '__main__':
unittest.main()