修复结束问题
This commit is contained in:
朱潮
parent
85a67aa1fa
commit
a451fd096d
@ -960,7 +960,20 @@ class OutputProcess:
|
||||
print(f"📥 输出进程收到结束信号")
|
||||
end_signal_received = True
|
||||
self.end_signal_received = True
|
||||
self.all_audio_received = True
|
||||
|
||||
# 只有在有音频数据或者TTS缓冲区有内容时，才设置all_audio_received
|
||||
# 这样可以避免在没有音频的情况下误判播放完成
|
||||
if (len(self.preload_buffer) > 0 or
|
||||
len(self.playback_buffer) > 0 or
|
||||
len(self.tts_buffer) > 0 or
|
||||
self.tts_task_queue.qsize() > 0):
|
||||
self.all_audio_received = True
|
||||
print(f"📥 设置all_audio_received=True，检测到有待处理的数据")
|
||||
else:
|
||||
# 如果没有任何数据，可能是LLM响应为空或其他问题
|
||||
# 也设置为True，让播放完成检测处理这种情况
|
||||
self.all_audio_received = True
|
||||
print(f"📥 警告：收到结束信号但没有检测到任何音频数据，可能存在问题")
|
||||
|
||||
# 重置完成事件标记
|
||||
self.completion_sent = False
|
||||
@ -991,6 +1004,14 @@ class OutputProcess:
|
||||
# 流式文本处理 - 智能缓冲
|
||||
streaming_text = audio_data[15:] # 移除 "STREAMING_TEXT:" 前缀
|
||||
print(f"📥 输出进程收到流式文本: {streaming_text}")
|
||||
|
||||
# 检查是否需要重置状态（新的对话开始）
|
||||
if self.end_signal_received:
|
||||
print(f"📥 检测到新对话开始，重置end_signal_received状态")
|
||||
self.end_signal_received = False
|
||||
self.all_audio_received = False
|
||||
self.completion_sent = False
|
||||
|
||||
self.process_streaming_text(streaming_text)
|
||||
continue
|
||||
|
||||
@ -998,6 +1019,14 @@ class OutputProcess:
|
||||
# 完整文本处理 - 强制刷新缓冲区
|
||||
complete_text = audio_data[14:] # 移除 "COMPLETE_TEXT:" 前缀
|
||||
print(f"📥 输出进程收到完整文本: {complete_text}")
|
||||
|
||||
# 检查是否需要重置状态（新的对话开始）
|
||||
if self.end_signal_received:
|
||||
print(f"📥 检测到新对话开始，重置end_signal_received状态")
|
||||
self.end_signal_received = False
|
||||
self.all_audio_received = False
|
||||
self.completion_sent = False
|
||||
|
||||
self.process_complete_text(complete_text)
|
||||
continue
|
||||
|
||||
@ -1011,6 +1040,10 @@ class OutputProcess:
|
||||
# LLM生成完成信号
|
||||
print(f"📥 输出进程收到LLM生成完成信号")
|
||||
self.llm_generation_complete = True
|
||||
# LLM完成后，如果还没有开始TTS，重置TTS完成状态
|
||||
if self.tts_generation_complete and self.tts_task_queue.qsize() == 0:
|
||||
print(f"📥 LLM完成但TTS队列为空，重置TTS完成状态为False")
|
||||
self.tts_generation_complete = False
|
||||
continue
|
||||
|
||||
if isinstance(audio_data, str) and audio_data.startswith("TTS_COMPLETE:"):
|
||||
@ -1024,6 +1057,11 @@ class OutputProcess:
|
||||
# 更新最后收到音频数据的时间
|
||||
self.last_audio_time = time.time()
|
||||
|
||||
# 如果之前已经收到结束信号，现在又收到音频数据，重置状态
|
||||
if self.all_audio_received:
|
||||
print(f"📥 收到新音频数据，重置all_audio_received状态为False")
|
||||
self.all_audio_received = False
|
||||
|
||||
# 减少日志输出，提高性能
|
||||
if processed_count <= 3 or processed_count % 50 == 0:
|
||||
print(f"📥 输出进程收到音频数据: {len(audio_data)} 字节")
|
||||
@ -1246,7 +1284,12 @@ class OutputProcess:
|
||||
time_since_last_chunk = 0
|
||||
print(f" - 最后播放时间: 尚未开始播放")
|
||||
|
||||
# 检查是否所有条件都满足 - 使用更新的状态变量
|
||||
# 检查TTS是否正在生成 - 新增条件
|
||||
# TTS正在生成的条件：队列中有任务 或 TTS生成未完成 或 还有待处理的缓冲区内容
|
||||
tts_is_generating = (tts_queue_size > 0 or not self.tts_generation_complete or len(self.tts_buffer) > 0)
|
||||
print(f" - TTS正在生成: {tts_is_generating} (队列:{tts_queue_size}, 完成:{self.tts_generation_complete}, 缓冲:{len(self.tts_buffer)})")
|
||||
|
||||
# 检查是否所有条件都满足 - 使用更新的状态变量，添加TTS生成状态检查
|
||||
all_conditions_met = (
|
||||
self.llm_generation_complete and
|
||||
self.tts_generation_complete and
|
||||
@ -1254,7 +1297,8 @@ class OutputProcess:
|
||||
self.pre_buffer_empty and
|
||||
self.playback_buffer_empty and
|
||||
self.no_active_playback and # 使用状态变量
|
||||
tts_queue_size == 0
|
||||
tts_queue_size == 0 and
|
||||
not tts_is_generating # 新增：确保TTS不在生成中
|
||||
)
|
||||
|
||||
if all_conditions_met:
|
||||
@ -1263,10 +1307,17 @@ class OutputProcess:
|
||||
print(f"✅ 所有播放完成条件已满足，且{time_since_last_chunk:.2f}秒无新音频，可以结束播放")
|
||||
return True
|
||||
elif self.last_audio_chunk_time == 0:
|
||||
# 如果从未开始播放，但有音频数据，说明播放可能有问题
|
||||
# 如果从未开始播放，检查是否有音频数据
|
||||
if len(self.playback_buffer) == 0 and len(self.preload_buffer) == 0:
|
||||
print(f"⚠️ 从未开始播放且无音频数据，可能播放失败，强制结束")
|
||||
return True
|
||||
# 检查是否有TTS任务在排队
|
||||
if self.tts_task_queue.qsize() == 0 and len(self.tts_buffer) == 0:
|
||||
# 真的没有任何数据，可能是LLM响应为空或TTS失败
|
||||
print(f"⚠️ 从未开始播放且无任何数据，可能LLM响应为空或TTS失败，强制结束")
|
||||
return True
|
||||
else:
|
||||
# 还有TTS任务待处理
|
||||
print(f"⏳ 从未开始播放但还有TTS任务待处理，等待TTS生成...")
|
||||
return False
|
||||
else:
|
||||
print(f"⏳ 从未开始播放但还有音频数据，等待播放开始...")
|
||||
return False
|
||||
@ -1278,7 +1329,8 @@ class OutputProcess:
|
||||
if (self.llm_generation_complete and
|
||||
self.tts_generation_complete and
|
||||
self.all_audio_received and
|
||||
tts_queue_size == 0):
|
||||
tts_queue_size == 0 and
|
||||
not tts_is_generating): # 新增：确保TTS不在生成中
|
||||
|
||||
if self.pre_buffer_empty and self.playback_buffer_empty:
|
||||
if self.no_active_playback: # 使用状态变量
|
||||
@ -1473,6 +1525,10 @@ class OutputProcess:
|
||||
self.is_playing = False
|
||||
self.completion_sent = False # 重置完成标记
|
||||
|
||||
# 重置TTS生成完成状态 - 关键修复
|
||||
self.tts_generation_complete = False
|
||||
print(f"🔊 已重置TTS生成完成状态为False，开始新的TTS生成")
|
||||
|
||||
# 构建请求头
|
||||
headers = {
|
||||
"X-Api-App-Id": self.tts_app_id,
|
||||
@ -1646,6 +1702,10 @@ class OutputProcess:
|
||||
combined_text = ''.join(self.tts_buffer)
|
||||
print(f"🔊 合并后的文本: '{combined_text}' (长度: {len(combined_text)})")
|
||||
|
||||
# 重置TTS生成完成状态 - 关键修复
|
||||
self.tts_generation_complete = False
|
||||
print(f"🔊 触发新TTS任务，已重置TTS生成完成状态为False")
|
||||
|
||||
# 添加到TTS任务队列
|
||||
print(f"🔊 尝试添加TTS任务到队列")
|
||||
if self._add_tts_task(combined_text):
|
||||
@ -1760,6 +1820,17 @@ class OutputProcess:
|
||||
print(f"🔊 过滤后的文本: '{filtered_text}' (长度: {len(filtered_text)})")
|
||||
|
||||
if filtered_text:
|
||||
# 检查是否是第一次收到流式文本（即所有状态都还是完成状态）
|
||||
if (self.llm_generation_complete and
|
||||
self.tts_generation_complete and
|
||||
self.all_audio_received):
|
||||
print(f"🔊 首次收到流式文本，重置所有播放完成检测状态")
|
||||
self.llm_generation_complete = False
|
||||
self.tts_generation_complete = False
|
||||
self.all_audio_received = False
|
||||
self.end_signal_received = False
|
||||
self.completion_sent = False
|
||||
|
||||
# 使用智能句子缓冲系统
|
||||
print(f"🔊 添加文本到智能缓冲区")
|
||||
self._add_sentence_to_buffer(filtered_text)
|
||||
@ -1775,6 +1846,14 @@ class OutputProcess:
|
||||
filtered_text = self._filter_parentheses_content(text.strip())
|
||||
|
||||
if filtered_text:
|
||||
# 重置所有播放完成检测状态 - 开始新的对话
|
||||
self.llm_generation_complete = False
|
||||
self.tts_generation_complete = False
|
||||
self.all_audio_received = False
|
||||
self.end_signal_received = False
|
||||
self.completion_sent = False
|
||||
print(f"🔊 处理完整文本：已重置所有播放完成检测状态")
|
||||
|
||||
# 直接添加到缓冲区并强制处理
|
||||
self.tts_buffer.append(filtered_text)
|
||||
self._process_tts_buffer()
|
||||
|
||||
@ -282,10 +282,16 @@ class ControlSystem:
|
||||
def _handle_idle_state(self):
|
||||
"""处理空闲状态"""
|
||||
if self.state == RecordingState.IDLE:
|
||||
# 启用输入进程录音功能
|
||||
self.input_command_queue.put(ControlCommand('enable_recording'))
|
||||
self.state = RecordingState.RECORDING
|
||||
print("🎯 状态：IDLE → RECORDING")
|
||||
# 检查是否应该立即启用录音
|
||||
# 如果刚刚从播放完成状态切换过来，不要立即启用录音
|
||||
if hasattr(self, '_just_finished_playing') and self._just_finished_playing:
|
||||
print("🎯 状态：IDLE（播放刚完成，等待延迟启用录音）")
|
||||
self._just_finished_playing = False # 重置标志
|
||||
else:
|
||||
# 启用输入进程录音功能
|
||||
self.input_command_queue.put(ControlCommand('enable_recording'))
|
||||
self.state = RecordingState.RECORDING
|
||||
print("🎯 状态：IDLE → RECORDING")
|
||||
|
||||
def _handle_recording_state(self):
|
||||
"""处理录音状态"""
|
||||
@ -362,24 +368,45 @@ class ControlSystem:
|
||||
self.stats['total_conversations'] += 1
|
||||
print(f"📡 主控制：已更新统计，对话数 = {self.stats['total_conversations']}")
|
||||
|
||||
# 等待一段时间确保音频设备完全停止播放
|
||||
print(f"📡 主控制：等待音频设备完全停止...")
|
||||
time.sleep(1.0) # 增加1秒等待时间
|
||||
|
||||
# 切换到空闲状态
|
||||
# 切换到空闲状态，但先不启用录音
|
||||
old_state = self.state.value
|
||||
self.state = RecordingState.IDLE
|
||||
# 设置标志，表示刚从播放状态切换过来
|
||||
self._just_finished_playing = True
|
||||
print(f"🎯 状态：{old_state} → IDLE")
|
||||
|
||||
# 重新启用输入进程录音功能
|
||||
try:
|
||||
self.input_command_queue.put(ControlCommand('enable_recording'))
|
||||
print(f"📡 主控制：已发送 enable_recording 命令到输入进程")
|
||||
print(f"📡 主控制：输入进程已重新启用，可以开始新的录音")
|
||||
except Exception as e:
|
||||
print(f"❌ 主控制：发送 enable_recording 命令失败: {e}")
|
||||
# 延迟重新启用录音，确保音频设备完全停止
|
||||
print(f"📡 主控制：延迟重新启用录音，避免录制到回声...")
|
||||
|
||||
print(f"📡 主控制：播放完成事件处理完成")
|
||||
# 延迟启用录音的函数
|
||||
def delayed_enable_recording():
|
||||
import threading
|
||||
import time
|
||||
|
||||
# 等待更长时间确保音频完全停止
|
||||
time.sleep(2.0) # 增加到2秒
|
||||
|
||||
# 检查输出队列是否还有音频数据
|
||||
output_queue_size = self.output_audio_queue.qsize()
|
||||
if output_queue_size > 0:
|
||||
print(f"📡 主控制：检测到输出队列仍有 {output_queue_size} 个音频块，继续等待...")
|
||||
time.sleep(1.0)
|
||||
|
||||
# 重新启用输入进程录音功能
|
||||
try:
|
||||
self.input_command_queue.put(ControlCommand('enable_recording'))
|
||||
print(f"📡 主控制：延迟发送 enable_recording 命令到输入进程")
|
||||
print(f"📡 主控制：输入进程已重新启用，可以开始新的录音")
|
||||
# 更新状态为录音状态
|
||||
self.state = RecordingState.RECORDING
|
||||
print(f"🎯 状态：IDLE → RECORDING（延迟启用）")
|
||||
except Exception as e:
|
||||
print(f"❌ 主控制：延迟发送 enable_recording 命令失败: {e}")
|
||||
|
||||
# 在新线程中延迟启用录音
|
||||
threading.Thread(target=delayed_enable_recording, daemon=True).start()
|
||||
|
||||
print(f"📡 主控制：播放完成事件处理完成，将在后台延迟启用录音")
|
||||
|
||||
def _process_audio_pipeline(self):
|
||||
"""处理音频流水线：STT + LLM + TTS"""
|
||||
@ -414,6 +441,16 @@ class ControlSystem:
|
||||
print("❌ 文本转语音失败")
|
||||
self._handle_processing_failure()
|
||||
return
|
||||
|
||||
# 在禁用LLM的情况下，也需要发送完成信号
|
||||
print("📡 主控制：禁用LLM模式下发送完成信号")
|
||||
# 由于LLM未启用，直接发送LLM完成信号
|
||||
self._notify_llm_complete()
|
||||
# 发送TTS完成信号
|
||||
tts_complete_command = "TTS_COMPLETE:"
|
||||
self.output_audio_queue.put(tts_complete_command)
|
||||
# 发送结束信号
|
||||
self.output_audio_queue.put(None)
|
||||
else:
|
||||
print("ℹ️ 文本转语音功能已禁用")
|
||||
# 直接发送结束信号
|
||||
@ -436,6 +473,21 @@ class ControlSystem:
|
||||
self.state = RecordingState.IDLE
|
||||
self.processing_complete = True
|
||||
self.playback_complete = True
|
||||
|
||||
# 发送完成信号，防止输出进程等待
|
||||
print("📡 主控制：失败处理模式下发送完成信号")
|
||||
try:
|
||||
# 发送LLM完成信号
|
||||
self._notify_llm_complete()
|
||||
# 发送TTS完成信号
|
||||
tts_complete_command = "TTS_COMPLETE:"
|
||||
self.output_audio_queue.put(tts_complete_command)
|
||||
# 发送结束信号
|
||||
self.output_audio_queue.put(None)
|
||||
print("📡 主控制：已发送失败处理完成信号")
|
||||
except Exception as e:
|
||||
print(f"❌ 发送失败处理信号失败: {e}")
|
||||
|
||||
print("🎯 状态：PROCESSING → IDLE (失败)")
|
||||
|
||||
def _speech_to_text(self, audio_data: bytes) -> Optional[str]:
|
||||
@ -970,6 +1022,12 @@ class ControlSystem:
|
||||
# 通知输出进程所有文本已发送完成
|
||||
self._flush_output_process_tts_buffer()
|
||||
|
||||
# 在流式模式下，只发送结束信号，不发送TTS完成信号
|
||||
# 让OutputProcess在真正完成TTS生成时自己设置TTS完成状态
|
||||
print("📡 主控制：流式模式下发送结束信号")
|
||||
self.output_audio_queue.put(None)
|
||||
print("📡 主控制：已发送流式模式结束信号")
|
||||
|
||||
return accumulated_text != ""
|
||||
|
||||
else:
|
||||
@ -1202,31 +1260,10 @@ class ControlSystem:
|
||||
print(f"\n✅ TTS音频生成完成: {chunk_count} 块, {total_audio_size / 1024:.1f} KB")
|
||||
print(f"📊 队列大小变化: {queue_size_before} -> {self.output_audio_queue.qsize()}")
|
||||
|
||||
# 等待音频数据被输出进程完全处理
|
||||
print(f"📦 TTS音频数据已全部发送，等待输出进程处理...")
|
||||
max_wait_time = 30 # 最多等待30秒
|
||||
wait_start_time = time.time()
|
||||
last_queue_size = self.output_audio_queue.qsize()
|
||||
|
||||
while time.time() - wait_start_time < max_wait_time:
|
||||
current_queue_size = self.output_audio_queue.qsize()
|
||||
|
||||
# 如果队列为空或队列大小不再变化，说明音频数据已被处理
|
||||
if current_queue_size == 0:
|
||||
print(f"✅ 音频队列已清空，可以发送结束信号")
|
||||
break
|
||||
elif current_queue_size == last_queue_size:
|
||||
# 队列大小不再变化，可能处理完成
|
||||
print(f"📦 队列大小稳定在 {current_queue_size}，等待确认...")
|
||||
time.sleep(1)
|
||||
if self.output_audio_queue.qsize() == current_queue_size:
|
||||
print(f"✅ 队列大小稳定，可以发送结束信号")
|
||||
break
|
||||
else:
|
||||
print(f"📦 等待队列处理: {current_queue_size} 个项目待处理")
|
||||
last_queue_size = current_queue_size
|
||||
|
||||
time.sleep(0.5)
|
||||
# 注意：TTS音频数据已经全部发送到队列，不需要等待队列清空
|
||||
# 立即发送结束信号，让OutputProcess的增强播放完成检测机制处理
|
||||
print(f"📦 TTS音频数据已全部发送到队列，立即发送结束信号")
|
||||
print(f"📦 让OutputProcess的增强播放完成检测机制来等待真正的播放完成")
|
||||
|
||||
# 发送TTS完成信号
|
||||
print(f"📦 发送TTS完成信号到输出进程")
|
||||
|
||||
Loading…
Reference in New Issue
Block a user