voiceagent/tests/test_audio_contract.py

"""PCM framing, per-direction metadata, and HTTP fragmentation guarantees."""

from contextlib import aclosing

import httpx
import pytest
from fastapi.testclient import TestClient
from pydantic import ValidationError

from voice_assistant.agent.mock import MockAgentAdapter
from voice_assistant.application.voice_session import VoiceSession
from voice_assistant.audio.models import AudioFormat, CANONICAL_AUDIO_FORMAT
from voice_assistant.audio_gateway.app import main
from voice_assistant.domain.session import SessionState
from voice_assistant.stt.mock import MockSTTAdapter
from voice_assistant.tts.mock import MockTTSAdapter
from voice_assistant.tts.voxcpm import VoxCPMTTSAdapter
from voice_assistant.turn_manager import TurnManager
from voice_assistant.voice_pipeline import VoicePipeline


@pytest.fixture
def anyio_backend():
    return "asyncio"


@pytest.mark.parametrize("field", ["sample_rate", "channels"])
@pytest.mark.parametrize("value", [0, -1, True, "16000", 1.5])
def test_audio_format_requires_positive_integer_dimensions(field, value):
    fields = {"encoding": "pcm_s16le", "sample_rate": 16000, "channels": 1}
    fields[field] = value
    with pytest.raises(ValidationError):
        AudioFormat(**fields)


def test_shared_format_cannot_be_mutated():
    with pytest.raises(ValidationError):
        CANONICAL_AUDIO_FORMAT.sample_rate = 48000
    assert CANONICAL_AUDIO_FORMAT.sample_rate == 16000


@pytest.mark.parametrize("channels", [1, 2])
@pytest.mark.parametrize("size", [0, 1, 2, 3, 4, 5, 8])
def test_pcm_validation_requires_complete_channel_frames(channels, size):
    format_ = AudioFormat(encoding="pcm_s16le", sample_rate=16000, channels=channels)
    if not size or size % (2 * channels):
        with pytest.raises(ValueError):
            format_.validate_chunk(bytes(size))
    else:
        format_.validate_chunk(bytes(size))


def make_pipeline(tts):
    return VoicePipeline(
        session=SessionState(), turn_manager=TurnManager(),
        agent=MockAgentAdapter(), tts=tts,
    )


@pytest.mark.anyio
async def test_invalid_input_does_not_advance_sequence_or_reach_stt():
    class STT(MockSTTAdapter):
        def __init__(self):
            super().__init__()
            self.received = []

        async def send_audio(self, audio):
            await super().send_audio(audio)
            self.received.append(audio)

    async def emit(event):
        pass

    stt = STT()
    session = VoiceSession(
        pipeline=make_pipeline(MockTTSAdapter()), create_stt=lambda: stt, emit=emit,
    )
    try:
        turn_id = await session.start_turn()
        with pytest.raises(ValueError, match="complete frames"):
            await session.send_audio(turn_id, 0, b"\x00")
        assert session.expected_audio_sequence == 0
        assert stt.received == []
        await session.send_audio(turn_id, 0, b"\x00\x00")
        assert session.expected_audio_sequence == 1
        assert stt.received == [b"\x00\x00"]
    finally:
        await session.aclose()


@pytest.mark.anyio
async def test_pipeline_rejects_misaligned_tts_output():
    class TTS(MockTTSAdapter):
        async def synthesize(self, text):
            yield b"\x00"

    async with aclosing(make_pipeline(TTS()).synthesize_response("Olá")) as stream:
        with pytest.raises(ValueError, match="complete frames"):
            await anext(stream)


class FragmentedStream(httpx.AsyncByteStream):
    def __init__(self, chunks):
        self.chunks = chunks

    async def __aiter__(self):
        for chunk in self.chunks:
            yield chunk


def mock_tts_http(monkeypatch, chunks, content_type="audio/pcm"):
    original_client = httpx.AsyncClient

    def handle(request):
        return httpx.Response(
            200, headers={"content-type": content_type},
            stream=FragmentedStream(chunks),
        )

    monkeypatch.setattr(
        httpx, "AsyncClient",
        lambda **kwargs: original_client(transport=httpx.MockTransport(handle), **kwargs),
    )


@pytest.mark.anyio
@pytest.mark.parametrize("chunks", [
    [b"\x01", b"\x02\x03", b"", b"\x04"],
    [b"\x01\x02\x03", b"\x04"],
    [b"\x01\x02\x03\x04"],
])
async def test_voxcpm_reassembles_http_fragments_without_losing_bytes(monkeypatch, chunks):
    mock_tts_http(monkeypatch, chunks)
    tts = VoxCPMTTSAdapter(server="http://test", model="test", voice="default")
    output = [chunk async for chunk in tts.synthesize("Olá")]
    assert b"".join(output) == b"\x01\x02\x03\x04"
    assert all(chunk and len(chunk) % 2 == 0 for chunk in output)


@pytest.mark.anyio
async def test_voxcpm_rejects_truncated_final_frame(monkeypatch):
    mock_tts_http(monkeypatch, [b"\x01\x02\x03"])
    tts = VoxCPMTTSAdapter(server="http://test", model="test", voice="default")
    with pytest.raises(ValueError, match="incomplete PCM frame"):
        _ = [chunk async for chunk in tts.synthesize("Olá")]


@pytest.mark.anyio
@pytest.mark.parametrize("content_type", ["audio/wav", "text/event-stream", "application/json"])
async def test_voxcpm_rejects_non_pcm_response(monkeypatch, content_type):
    mock_tts_http(monkeypatch, [b"data"], content_type)
    tts = VoxCPMTTSAdapter(server="http://test", model="test", voice="default")
    with pytest.raises(ValueError, match="raw PCM"):
        _ = [chunk async for chunk in tts.synthesize("Olá")]


@pytest.mark.parametrize("options", [{"response_format": "wav"}, {"stream_format": "sse"}])
def test_voxcpm_rejects_incompatible_transport_settings(options):
    with pytest.raises(ValueError):
        VoxCPMTTSAdapter(server="http://test", model="test", voice="default", **options)


def test_gateway_announces_provider_output_format_and_preserves_input(monkeypatch):
    class TTS(MockTTSAdapter):
        output_format = AudioFormat(encoding="pcm_s16le", sample_rate=24000, channels=1)

    monkeypatch.setattr(main, "create_tts_adapter", lambda **kwargs: TTS())
    monkeypatch.setattr(main, "create_agent_adapter", lambda **kwargs: MockAgentAdapter())
    with TestClient(main.app).websocket_connect("/ws/audio") as socket:
        socket.send_json({"type": "session.start"})
        ready = socket.receive_json()
        assert ready["audio"]["sample_rate"] == 16000
        assert ready["output_audio"] == {
            "encoding": "pcm_s16le", "sample_rate": 24000, "channels": 1,
        }