dograh/api/tests/test_ultravox_realtime_wrapper.py
Abhishek 01acf6ac30
fix: fix speech to speech model transitions (#545)
* fix: fix transition logic for realtime providers

* chore: run formatter

* chore: generate SDK and fix other realtime providers

* fix: fix ultravox node transitions
2026-07-15 18:36:36 +05:30

437 lines
13 KiB
Python

import json
from types import SimpleNamespace
from unittest.mock import AsyncMock, call
import pytest
from pipecat.adapters.schemas.function_schema import FunctionSchema
from pipecat.adapters.schemas.tools_schema import ToolsSchema
from pipecat.frames.frames import LLMMessagesAppendFrame, TTSSpeakFrame
from pipecat.processors.aggregators.llm_context import LLMContext
from pipecat.processors.frame_processor import FrameDirection
from websockets.exceptions import ConnectionClosedError
from websockets.frames import Close
from api.schemas.ai_model_configuration import EffectiveAIModelConfiguration
from api.services.configuration.registry import UltravoxRealtimeLLMConfiguration
from api.services.pipecat.realtime.ultravox_realtime import (
DograhUltravoxOneShotInputParams,
DograhUltravoxRealtimeLLMService,
)
from api.services.pipecat.service_factory import create_realtime_llm_service
class _ClosingSocket:
def __init__(self, exc):
self._exc = exc
def __aiter__(self):
return self
async def __anext__(self):
raise self._exc
class _MessageSocket:
def __init__(self, messages):
self._messages = iter(messages)
def __aiter__(self):
return self
async def __anext__(self):
try:
return next(self._messages)
except StopIteration:
raise StopAsyncIteration
def _make_service() -> DograhUltravoxRealtimeLLMService:
service = DograhUltravoxRealtimeLLMService(
params=DograhUltravoxOneShotInputParams(
api_key="test-key",
model="ultravox-v0.7",
output_medium="voice",
),
settings=DograhUltravoxRealtimeLLMService.Settings(
model="ultravox-v0.7",
output_medium="voice",
),
)
service.stop_all_metrics = AsyncMock()
service.cancel_task = AsyncMock()
service.push_error = AsyncMock()
return service
def _tool_schema() -> ToolsSchema:
return ToolsSchema(
standard_tools=[
FunctionSchema(
name="transition_to_next_node",
description="Move to the next workflow node",
properties={"reason": {"type": "string"}},
required=[],
)
]
)
@pytest.mark.asyncio
async def test_tts_greeting_triggers_initial_connect():
service = _make_service()
service._connect_call = AsyncMock()
await service.process_frame(
TTSSpeakFrame("Hello there", append_to_context=True),
FrameDirection.DOWNSTREAM,
)
service._connect_call.assert_awaited_once()
assert service._connect_call.await_args.kwargs["greeting_text"] == "Hello there"
assert service._connect_call.await_args.kwargs["agent_speaks_first"] is True
@pytest.mark.asyncio
async def test_initial_context_connects_without_replay():
service = _make_service()
service._connect_call = AsyncMock()
context = LLMContext()
await service._handle_context(context)
service._connect_call.assert_awaited_once()
assert service._connect_call.await_args.kwargs["greeting_text"] is None
assert service._connect_call.await_args.kwargs["agent_speaks_first"] is True
@pytest.mark.asyncio
async def test_system_instruction_update_marks_stage_update_required():
service = _make_service()
service._socket = object()
changed = await service._update_settings(
DograhUltravoxRealtimeLLMService.Settings(system_instruction="new instruction")
)
assert "system_instruction" in changed
assert service._stage_update_required is True
@pytest.mark.asyncio
async def test_node_transition_updates_native_stage_without_reconnecting():
service = _make_service()
service._socket = object()
service._send = AsyncMock()
service._connect_call = AsyncMock()
service._pending_node_transition_tool_call_ids.add("call-transition")
service._stage_update_required = True
service._settings.system_instruction = "new instruction"
context = LLMContext(
messages=[
{
"role": "tool",
"tool_call_id": "call-transition",
"content": '{"status":"done"}',
},
],
tools=_tool_schema(),
)
await service._handle_context(context)
service._connect_call.assert_not_awaited()
service._send.assert_awaited_once()
message = service._send.await_args.args[0]
assert message["type"] == "client_tool_result"
assert message["invocationId"] == "call-transition"
assert message["responseType"] == "new-stage"
stage = json.loads(message["result"])
assert stage["systemPrompt"] == "new instruction"
assert stage["toolResultText"] == '{"status":"done"}'
assert stage["selectedTools"][0]["temporaryTool"]["modelToolName"] == (
"transition_to_next_node"
)
assert "call-transition" in service._completed_tool_calls
assert service._pending_node_transition_tool_call_ids == set()
assert service._stage_update_required is False
@pytest.mark.asyncio
async def test_ordinary_tool_result_uses_standard_tool_response():
service = _make_service()
service._socket = object()
service._send = AsyncMock()
context = LLMContext(
messages=[
{
"role": "tool",
"tool_call_id": "call-transition",
"content": '{"status":"done"}',
},
],
tools=_tool_schema(),
)
await service._handle_context(context)
service._send.assert_awaited_once_with(
{
"type": "client_tool_result",
"invocationId": "call-transition",
"result": '{"status":"done"}',
}
)
@pytest.mark.asyncio
async def test_only_registered_node_transition_invocations_are_tracked():
service = _make_service()
service.run_function_calls = AsyncMock()
service.register_function(
"transition_to_next_node",
AsyncMock(),
is_node_transition=True,
)
await service._handle_tool_invocation(
"transition_to_next_node", "call-transition", {"reason": "pricing"}
)
await service._handle_tool_invocation("lookup_price", "call-lookup", {})
assert service._pending_node_transition_tool_call_ids == {"call-transition"}
assert service.run_function_calls.await_count == 2
@pytest.mark.asyncio
async def test_node_transition_invocation_waits_for_response_end():
service = _make_service()
service.run_function_calls = AsyncMock()
service.stop_processing_metrics = AsyncMock()
service.push_frame = AsyncMock()
service.register_function(
"transition_to_next_node",
AsyncMock(),
is_node_transition=True,
)
service._bot_responding = "voice"
await service._handle_tool_invocation(
"transition_to_next_node", "call-transition", {"reason": "pricing"}
)
service.run_function_calls.assert_not_awaited()
assert service._deferred_node_transition_tool_invocations == [
(
"transition_to_next_node",
"call-transition",
{"reason": "pricing"},
)
]
await service._handle_response_end()
service.run_function_calls.assert_awaited_once()
function_call = service.run_function_calls.await_args.args[0][0]
assert function_call.function_name == "transition_to_next_node"
assert function_call.tool_call_id == "call-transition"
assert function_call.arguments == {"reason": "pricing"}
assert service._deferred_node_transition_tool_invocations == []
assert service._bot_responding is None
@pytest.mark.asyncio
async def test_ordinary_tool_invocation_runs_while_response_is_active():
service = _make_service()
service.run_function_calls = AsyncMock()
service._bot_responding = "voice"
await service._handle_tool_invocation("lookup_price", "call-lookup", {})
service.run_function_calls.assert_awaited_once()
assert service._deferred_node_transition_tool_invocations == []
def test_ultravox_requires_transition_context_aggregation():
service = _make_service()
assert service._requires_node_transition_context_aggregation() is True
@pytest.mark.asyncio
async def test_messages_append_frame_sends_user_text():
service = _make_service()
service._socket = object()
service._call_started = True
service._send_user_text = AsyncMock()
await service._handle_messages_append(
LLMMessagesAppendFrame(
[{"role": "user", "content": "Are you still there?"}],
run_llm=True,
)
)
service._send_user_text.assert_awaited_once_with("Are you still there?")
@pytest.mark.asyncio
async def test_messages_append_frame_queues_user_text_until_call_started():
service = _make_service()
service._socket = object()
service._call_started = False
service._send_user_text = AsyncMock()
await service._handle_messages_append(
LLMMessagesAppendFrame(
[{"role": "user", "content": "Are you still there?"}],
run_llm=True,
)
)
assert service._pending_user_text_messages == ["Are you still there?"]
service._send_user_text.assert_not_awaited()
@pytest.mark.asyncio
async def test_call_started_flushes_pending_user_text_messages():
service = _make_service()
service._pending_user_text_messages = [
"First queued message",
"Second queued message",
]
service._send_user_text = AsyncMock()
service._socket = _MessageSocket(['{"type":"call_started","callId":"call-123"}'])
await service._receive_messages()
assert service._call_started is True
assert service._pending_user_text_messages == []
assert service._send_user_text.await_args_list == [
call("First queued message"),
call("Second queued message"),
]
@pytest.mark.asyncio
async def test_completed_input_transcription_is_broadcast_as_finalized():
service = _make_service()
service.broadcast_frame = AsyncMock()
service._last_user_id = "caller-1"
await service._handle_user_transcript("Hello there")
service.broadcast_frame.assert_awaited_once()
assert service.broadcast_frame.await_args.args[0].__name__ == "TranscriptionFrame"
assert service.broadcast_frame.await_args.kwargs["text"] == "Hello there"
assert service.broadcast_frame.await_args.kwargs["finalized"] is True
def test_build_one_shot_params_uses_explicit_greeting_text():
service = _make_service()
params = service._build_one_shot_params(
greeting_text="Welcome to Dograh",
agent_speaks_first=True,
)
assert params.extra["firstSpeakerSettings"] == {
"agent": {"text": "Welcome to Dograh"}
}
def test_build_one_shot_params_uses_current_system_instruction():
service = _make_service()
service._settings.system_instruction = "Base instruction"
params = service._build_one_shot_params(
greeting_text=None,
agent_speaks_first=True,
)
assert params.system_prompt == "Base instruction"
def test_to_selected_tools_includes_registered_timeout():
service = _make_service()
service.register_function(
"transition_to_next_node",
AsyncMock(),
timeout_secs=5.5,
)
selected_tools = service._to_selected_tools(_tool_schema())
assert selected_tools == [
{
"temporaryTool": {
"modelToolName": "transition_to_next_node",
"description": "Move to the next workflow node",
"dynamicParameters": [
{
"name": "reason",
"location": "PARAMETER_LOCATION_BODY",
"schema": {"type": "string"},
"required": False,
}
],
"client": {},
"timeout": "5.5s",
}
}
]
@pytest.mark.asyncio
async def test_receive_messages_ignores_benign_websocket_close():
service = _make_service()
service._socket = _ClosingSocket(
ConnectionClosedError(None, Close(1000, "OK"), None)
)
await service._receive_messages()
service.push_error.assert_not_awaited()
@pytest.mark.asyncio
async def test_receive_messages_reports_unexpected_websocket_close():
service = _make_service()
service._socket = _ClosingSocket(
ConnectionClosedError(None, Close(1011, "internal error"), None)
)
await service._receive_messages()
service.push_error.assert_awaited_once()
def test_factory_creates_dograh_ultravox_realtime_service():
effective_config = EffectiveAIModelConfiguration(
is_realtime=True,
realtime=UltravoxRealtimeLLMConfiguration(
provider="ultravox_realtime",
api_key="ultra-key",
model="ultravox-v0.7",
voice="Mark",
),
)
service = create_realtime_llm_service(
effective_config,
audio_config=SimpleNamespace(),
)
assert isinstance(service, DograhUltravoxRealtimeLLMService)
assert service._params.voice == "Mark"
def test_ultravox_realtime_configuration_defaults_to_mark_voice():
config = UltravoxRealtimeLLMConfiguration(
provider="ultravox_realtime",
api_key="ultra-key",
model="ultravox-v0.7",
)
assert config.voice == "Mark"