LLM_Avatar / src /open_llm_vtuber /conversations /conversation_handler.py
andry
initial commit: customized LLM-VTuber setup without binary files
f90aaa2
Raw
History Blame Contribute Delete
7.96 kB
import asyncio
import json
from typing import Dict, Optional, Callable
import numpy as np
from fastapi import WebSocket
from loguru import logger
from ..chat_group import ChatGroupManager
from ..chat_history_manager import store_message
from ..service_context import ServiceContext
from .group_conversation import process_group_conversation
from .single_conversation import process_single_conversation
from .conversation_utils import EMOJI_LIST
from .types import GroupConversationState
from prompts import prompt_loader
async def handle_conversation_trigger(
msg_type: str,
data: dict,
client_uid: str,
context: ServiceContext,
websocket: WebSocket,
client_contexts: Dict[str, ServiceContext],
client_connections: Dict[str, WebSocket],
chat_group_manager: ChatGroupManager,
received_data_buffers: Dict[str, np.ndarray],
current_conversation_tasks: Dict[str, Optional[asyncio.Task]],
broadcast_to_group: Callable,
) -> None:
"""Handle triggers that start a conversation"""
metadata = None
if msg_type == "ai-speak-signal":
try:
# Get proactive speak prompt from config
prompt_name = "proactive_speak_prompt"
prompt_file = context.system_config.tool_prompts.get(prompt_name)
if prompt_file:
user_input = prompt_loader.load_util(prompt_file)
else:
logger.warning("Proactive speak prompt not configured, using default")
user_input = "Please say something."
except Exception as e:
logger.error(f"Error loading proactive speak prompt: {e}")
user_input = "Please say something."
# Add metadata to indicate this is a proactive speak request
# that should be skipped in both memory and history
metadata = {
"proactive_speak": True,
"skip_memory": True, # Skip storing in AI's internal memory
"skip_history": True, # Skip storing in local conversation history
}
await websocket.send_text(
json.dumps(
{
"type": "full-text",
"text": "AI wants to speak something...",
}
)
)
elif msg_type == "text-input":
user_input = data.get("text", "")
else: # mic-audio-end
user_input = received_data_buffers[client_uid]
received_data_buffers[client_uid] = np.array([])
images = data.get("images")
session_emoji = np.random.choice(EMOJI_LIST)
group = chat_group_manager.get_client_group(client_uid)
if group and len(group.members) > 1:
# Use group_id as task key for group conversations
task_key = group.group_id
if (
task_key not in current_conversation_tasks
or current_conversation_tasks[task_key].done()
):
logger.info(f"Starting new group conversation for {task_key}")
current_conversation_tasks[task_key] = asyncio.create_task(
process_group_conversation(
client_contexts=client_contexts,
client_connections=client_connections,
broadcast_func=broadcast_to_group,
group_members=group.members,
initiator_client_uid=client_uid,
user_input=user_input,
images=images,
session_emoji=session_emoji,
metadata=metadata,
)
)
else:
# Use client_uid as task key for individual conversations
current_conversation_tasks[client_uid] = asyncio.create_task(
process_single_conversation(
context=context,
websocket_send=websocket.send_text,
client_uid=client_uid,
user_input=user_input,
images=images,
session_emoji=session_emoji,
metadata=metadata,
)
)
async def handle_individual_interrupt(
client_uid: str,
current_conversation_tasks: Dict[str, Optional[asyncio.Task]],
context: ServiceContext,
heard_response: str,
):
if client_uid in current_conversation_tasks:
task = current_conversation_tasks[client_uid]
if task and not task.done():
task.cancel()
logger.info("🛑 Conversation task was successfully interrupted")
try:
context.agent_engine.handle_interrupt(heard_response)
except Exception as e:
logger.error(f"Error handling interrupt: {e}")
if context.history_uid:
store_message(
conf_uid=context.character_config.conf_uid,
history_uid=context.history_uid,
role="ai",
content=heard_response,
name=context.character_config.character_name,
avatar=context.character_config.avatar,
)
store_message(
conf_uid=context.character_config.conf_uid,
history_uid=context.history_uid,
role="system",
content="[Interrupted by user]",
)
async def handle_group_interrupt(
group_id: str,
heard_response: str,
current_conversation_tasks: Dict[str, Optional[asyncio.Task]],
chat_group_manager: ChatGroupManager,
client_contexts: Dict[str, ServiceContext],
broadcast_to_group: Callable,
) -> None:
"""Handles interruption for a group conversation"""
task = current_conversation_tasks.get(group_id)
if not task or task.done():
return
# Get state and speaker info before cancellation
state = GroupConversationState.get_state(group_id)
current_speaker_uid = state.current_speaker_uid if state else None
# Get context from current speaker
context = None
group = chat_group_manager.get_group_by_id(group_id)
if current_speaker_uid:
context = client_contexts.get(current_speaker_uid)
logger.info(f"Found current speaker context for {current_speaker_uid}")
if not context and group and group.members:
logger.warning(f"No context found for group {group_id}, using first member")
context = client_contexts.get(next(iter(group.members)))
# Now cancel the task
task.cancel()
try:
await task
except asyncio.CancelledError:
logger.info(f"🛑 Group conversation {group_id} cancelled successfully.")
current_conversation_tasks.pop(group_id, None)
GroupConversationState.remove_state(group_id) # Clean up state after we've used it
# Store messages with speaker info
if context and group:
for member_uid in group.members:
if member_uid in client_contexts:
try:
member_ctx = client_contexts[member_uid]
member_ctx.agent_engine.handle_interrupt(heard_response)
store_message(
conf_uid=member_ctx.character_config.conf_uid,
history_uid=member_ctx.history_uid,
role="ai",
content=heard_response,
name=context.character_config.character_name,
avatar=context.character_config.avatar,
)
store_message(
conf_uid=member_ctx.character_config.conf_uid,
history_uid=member_ctx.history_uid,
role="system",
content="[Interrupted by user]",
)
except Exception as e:
logger.error(f"Error handling interrupt for {member_uid}: {e}")
await broadcast_to_group(
list(group.members),
{
"type": "interrupt-signal",
"text": "conversation-interrupted",
},
)