Skip to content

Commit 5813d84

Browse files
fix(voice): preserve legacy positional config arguments (#5198)
1 parent 4658a0e commit 5813d84

3 files changed

Lines changed: 174 additions & 1 deletion

File tree

‎src/agents/voice/pipeline_config.py‎

Lines changed: 40 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,8 @@
11
from __future__ import annotations
22

33
from dataclasses import dataclass, field
4-
from typing import TYPE_CHECKING, Any
4+
from functools import wraps
5+
from typing import TYPE_CHECKING, Any, overload
56

67
from .._config_coercion import _declared_dataclass_type, coerce_dataclass_config
78
from ..tracing import TracingConfig
@@ -10,6 +11,24 @@
1011
from .models.openai_model_provider import OpenAIVoiceModelProvider
1112

1213

14+
def _preserve_legacy_positional_arguments(
15+
cls: type[VoicePipelineConfig],
16+
) -> type[VoicePipelineConfig]:
17+
generated_init = cls.__init__
18+
19+
@wraps(generated_init)
20+
def init(self: VoicePipelineConfig, *args: Any, **kwargs: Any) -> None:
21+
# Before tracing was added, the third positional argument was the text privacy flag.
22+
# Insert only that slot so the dataclass still owns defaults and argument validation.
23+
if len(args) >= 3 and isinstance(args[2], bool):
24+
args = (*args[:2], kwargs.pop("tracing", None), *args[2:])
25+
generated_init(self, *args, **kwargs)
26+
27+
cls.__init__ = init # type: ignore[method-assign]
28+
return cls
29+
30+
31+
@_preserve_legacy_positional_arguments
1332
@dataclass
1433
class VoicePipelineConfig:
1534
"""Configuration for a `VoicePipeline`."""
@@ -54,6 +73,7 @@ class VoicePipelineConfig:
5473

5574
if TYPE_CHECKING:
5675

76+
@overload
5777
def __init__(
5878
self,
5979
model_provider: VoiceModelProvider = ...,
@@ -68,6 +88,25 @@ def __init__(
6888
tts_settings: TTSModelSettings | dict[str, Any] = ...,
6989
) -> None: ...
7090

91+
@overload
92+
def __init__(
93+
self,
94+
model_provider: VoiceModelProvider,
95+
tracing_disabled: bool,
96+
trace_include_sensitive_data: bool,
97+
/,
98+
trace_include_sensitive_audio_data: bool = True,
99+
workflow_name: str = "Voice Agent",
100+
group_id: str = ...,
101+
trace_metadata: dict[str, Any] | None = None,
102+
stt_settings: STTModelSettings | dict[str, Any] = ...,
103+
tts_settings: TTSModelSettings | dict[str, Any] = ...,
104+
*,
105+
tracing: TracingConfig | None = None,
106+
) -> None: ...
107+
108+
def __init__(self, *args: Any, **kwargs: Any) -> None: ...
109+
71110
def __post_init__(self) -> None:
72111
self.stt_settings = coerce_dataclass_config(
73112
self.stt_settings,
Lines changed: 83 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,83 @@
1+
# pyright: reportCallIssue=true, reportArgumentType=true
2+
from dataclasses import replace
3+
from typing import Any
4+
5+
import pytest
6+
7+
from agents.tracing import TracingConfig
8+
from agents.voice import STTModelSettings, TTSModelSettings, VoicePipelineConfig
9+
from agents.voice.models.openai_model_provider import OpenAIVoiceModelProvider
10+
11+
12+
@pytest.mark.parametrize("legacy", [False, True])
13+
def test_released_positional_config_fields(legacy: bool) -> None:
14+
provider = OpenAIVoiceModelProvider()
15+
tracing = TracingConfig(api_key="synthetic-placeholder")
16+
metadata = {"source": "synthetic"}
17+
stt = STTModelSettings(language="en")
18+
tts = TTSModelSettings(voice="alloy")
19+
if legacy:
20+
config = VoicePipelineConfig(
21+
provider,
22+
False,
23+
False,
24+
False,
25+
"workflow",
26+
"group",
27+
metadata,
28+
stt,
29+
tts,
30+
tracing=tracing,
31+
)
32+
else:
33+
config = VoicePipelineConfig(
34+
provider, False, tracing, False, False, "workflow", "group", metadata, stt, tts
35+
)
36+
37+
assert config.model_provider is provider
38+
assert config.tracing_disabled is False
39+
assert config.tracing is tracing
40+
assert config.trace_include_sensitive_data is False
41+
assert config.trace_include_sensitive_audio_data is False
42+
assert config.workflow_name == "workflow"
43+
assert config.group_id == "group"
44+
assert config.trace_metadata is metadata
45+
assert config.stt_settings is stt
46+
assert config.tts_settings is tts
47+
assert replace(config, workflow_name="replacement") == VoicePipelineConfig(
48+
provider, False, tracing, False, False, "replacement", "group", metadata, stt, tts
49+
)
50+
51+
52+
def test_legacy_positional_config_keeps_defaults_and_keyword_settings() -> None:
53+
provider = OpenAIVoiceModelProvider()
54+
config = VoicePipelineConfig(provider, False, False)
55+
assert config.tracing is None
56+
assert config.trace_include_sensitive_data is False
57+
assert config.trace_include_sensitive_audio_data is True
58+
assert config.workflow_name == "Voice Agent"
59+
assert config.group_id.startswith("group_")
60+
assert config.trace_metadata is None
61+
assert config.stt_settings == STTModelSettings()
62+
assert config.tts_settings == TTSModelSettings()
63+
64+
config = VoicePipelineConfig(
65+
provider,
66+
True,
67+
True,
68+
trace_include_sensitive_audio_data=False,
69+
stt_settings={"language": "en"},
70+
tts_settings={"voice": "alloy"},
71+
)
72+
assert config.tracing_disabled is True
73+
assert config.tracing is None
74+
assert config.trace_include_sensitive_data is True
75+
assert config.trace_include_sensitive_audio_data is False
76+
assert config.stt_settings.language == "en"
77+
assert config.tts_settings.voice == "alloy"
78+
79+
80+
def test_legacy_positional_config_rejects_duplicate_arguments() -> None:
81+
conflicting: dict[str, Any] = {"trace_include_sensitive_data": True}
82+
with pytest.raises(TypeError, match="multiple values.*trace_include_sensitive_data"):
83+
VoicePipelineConfig(OpenAIVoiceModelProvider(), False, False, **conflicting)

‎tests/voice/test_pipeline_tracing.py‎

Lines changed: 51 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -29,6 +29,7 @@
2929
VoicePipelineConfig,
3030
VoiceWorkflowBase,
3131
)
32+
from agents.voice.models.openai_model_provider import OpenAIVoiceModelProvider
3233
from agents.voice.models.openai_stt import OpenAISTTModel
3334
from agents.voice.models.openai_tts import OpenAITTSModel
3435
from tests.testing_processor import (
@@ -151,6 +152,56 @@ def _export_payload(monkeypatch: pytest.MonkeyPatch) -> list[dict[str, Any]]:
151152
return http_client.post.call_args.kwargs["json"]["data"] if http_client.post.called else []
152153

153154

155+
@pytest.mark.asyncio
156+
@pytest.mark.parametrize("streamed", [False, True])
157+
@pytest.mark.parametrize("include_audio", [False, True])
158+
@pytest.mark.parametrize("legacy", [False, True])
159+
async def test_released_positional_privacy_controls_exported_voice_payloads(
160+
streamed: bool,
161+
include_audio: bool,
162+
legacy: bool,
163+
speech_client: tuple[Any, ...],
164+
monkeypatch: pytest.MonkeyPatch,
165+
) -> None:
166+
client, transcribe, speech, websocket = speech_client
167+
provider = OpenAIVoiceModelProvider()
168+
config = (
169+
VoicePipelineConfig(provider, False, False, include_audio)
170+
if legacy
171+
else VoicePipelineConfig(provider, False, None, False, include_audio)
172+
)
173+
result = await _pipeline(client, config).run(await _input(streamed))
174+
events = [event async for event in result.stream()]
175+
assert result.total_output_text == _TEXT
176+
assert (
177+
b"".join(
178+
cast(np.ndarray[Any, Any], event.data).tobytes()
179+
for event in events
180+
if event.type == "voice_stream_event_audio"
181+
)
182+
== _PCM.tobytes()
183+
)
184+
assert speech.call_args.kwargs["input"] == _TEXT
185+
if streamed:
186+
assert websocket.sent[1]["audio"] == _ENCODED
187+
assert websocket.closed
188+
else:
189+
assert transcribe.call_args.kwargs["file"][1].getvalue().endswith(_PCM.tobytes())
190+
191+
voice_spans = [
192+
item["span_data"]
193+
for item in _export_payload(monkeypatch)
194+
if item["object"] == "trace.span"
195+
and item["span_data"]["type"] in {"transcription", "speech"}
196+
]
197+
assert {span["type"] for span in voice_spans} == {"transcription", "speech"}
198+
for kind in ("transcription", "speech"):
199+
payload = json.dumps([span for span in voice_spans if span["type"] == kind])
200+
assert (_ENCODED in payload) is include_audio
201+
assert _TRANSCRIPT not in payload
202+
assert _TEXT not in payload
203+
204+
154205
@pytest.mark.asyncio
155206
@pytest.mark.parametrize("streamed", [False, True])
156207
@pytest.mark.parametrize(

0 commit comments

Comments
 (0)