test: complete VAD unit and integration tests

Including an integration test with real voice audio.

ref: N25B-213

test/integration/agents/vad_agent/test_vad_agent.py

@@ -0,0 +1,97 @@
+from unittest.mock import MagicMock, AsyncMock, patch
+
+import pytest
+import zmq
+from spade.agent import Agent
+
+from control_backend.agents.vad_agent import VADAgent
+
+
+@pytest.fixture
+def zmq_context(mocker):
+    return mocker.patch("control_backend.agents.vad_agent.zmq_context")
+
+
+@pytest.fixture
+def streaming(mocker):
+    return mocker.patch("control_backend.agents.vad_agent.Streaming")
+
+
+@pytest.mark.asyncio
+async def test_normal_setup(streaming):
+    """
+    Test that during normal setup, the VAD agent creates a Streaming behavior and creates audio sockets.
+    """
+    vad_agent = VADAgent("tcp://localhost:12345", False)
+    vad_agent.add_behaviour = MagicMock()
+
+    await vad_agent.setup()
+
+    streaming.assert_called_once()
+    vad_agent.add_behaviour.assert_called_once_with(streaming.return_value)
+    assert vad_agent.audio_in_socket is not None
+    assert vad_agent.audio_out_socket is not None
+
+
+@pytest.mark.parametrize("do_bind", [True, False])
+def test_in_socket_creation(zmq_context, do_bind: bool):
+    """
+    Test that the VAD agent creates an audio input socket, differentiating between binding and connecting.
+    """
+    vad_agent = VADAgent(f"tcp://{"*" if do_bind else "localhost"}:12345", do_bind)
+
+    vad_agent._connect_audio_in_socket()
+
+    assert vad_agent.audio_in_socket is not None
+
+    zmq_context.socket.assert_called_once_with(zmq.SUB)
+    zmq_context.socket.return_value.setsockopt_string.assert_called_once_with(zmq.SUBSCRIBE, "")
+
+    if do_bind:
+        zmq_context.socket.return_value.bind.assert_called_once_with("tcp://*:12345")
+    else:
+        zmq_context.socket.return_value.connect.assert_called_once_with("tcp://localhost:12345")
+
+
+def test_out_socket_creation(zmq_context):
+    """
+    Test that the VAD agent creates an audio output socket correctly.
+    """
+    vad_agent = VADAgent("tcp://localhost:12345", False)
+
+    vad_agent._connect_audio_out_socket()
+
+    assert vad_agent.audio_out_socket is not None
+
+    zmq_context.socket.assert_called_once_with(zmq.PUB)
+    zmq_context.socket.return_value.bind_to_random_port.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_out_socket_creation_failure(zmq_context):
+    """
+    Test setup failure when the audio output socket cannot be created.
+    """
+    with patch.object(Agent, "stop", new_callable=AsyncMock) as mock_super_stop:
+        zmq_context.socket.return_value.bind_to_random_port.side_effect = zmq.ZMQBindError
+        vad_agent = VADAgent("tcp://localhost:12345", False)
+
+        await vad_agent.setup()
+
+        assert vad_agent.audio_out_socket is None
+        mock_super_stop.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_stop(zmq_context):
+    """
+    Test that when the VAD agent is stopped, the sockets are closed correctly.
+    """
+    vad_agent = VADAgent("tcp://localhost:12345", False)
+
+    await vad_agent.setup()
+    await vad_agent.stop()
+
+    assert zmq_context.socket.return_value.close.call_count == 2
+    assert vad_agent.audio_in_socket is None
+    assert vad_agent.audio_out_socket is None

test/integration/agents/vad_agent/test_vad_with_audio.py

@@ -0,0 +1,57 @@
+import os
+from unittest.mock import MagicMock, AsyncMock
+
+import pytest
+import soundfile as sf
+import zmq
+
+from control_backend.agents.vad_agent import Streaming
+
+
+def get_audio_chunks() -> list[bytes]:
+    curr_file = os.path.realpath(__file__)
+    curr_dir = os.path.dirname(curr_file)
+    file = f"{curr_dir}/speech_with_pauses_16k_1c_float32.wav"
+
+    chunk_size = 512
+
+    chunks = []
+
+    with sf.SoundFile(file, 'r') as f:
+        assert f.samplerate == 16000
+        assert f.channels == 1
+        assert f.subtype == "FLOAT"
+
+        while True:
+            data = f.read(chunk_size, dtype="float32")
+            if len(data) != chunk_size:
+                break
+
+            chunks.append(data.tobytes())
+
+    return chunks
+
+
+@pytest.mark.asyncio
+async def test_real_audio(mocker):
+    """
+    Test the VAD agent with only input and output mocked. Using the real model, using real audio as
+    input. Ensure that it outputs some fragments with audio.
+    """
+    audio_chunks = get_audio_chunks()
+    audio_in_socket = AsyncMock()
+    audio_in_socket.recv.side_effect = audio_chunks
+
+    mock_poller: MagicMock = mocker.patch("control_backend.agents.vad_agent.zmq.Poller")
+    mock_poller.return_value.poll.return_value = [(audio_in_socket, zmq.POLLIN)]
+
+    audio_out_socket = AsyncMock()
+
+    vad_streamer = Streaming(audio_in_socket, audio_out_socket)
+    for _ in audio_chunks:
+        await vad_streamer.run()
+
+    audio_out_socket.send.assert_called()
+    for args in audio_out_socket.send.call_args_list:
+        assert isinstance(args[0][0], bytes)
+        assert len(args[0][0]) >= 512*4*3  # Should be at least 3 chunks of audio

test/unit/agents/test_vad_socket_poller.py

@@ -0,0 +1,46 @@
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+import zmq
+
+from control_backend.agents.vad_agent import SocketPoller
+
+
+@pytest.fixture
+def socket():
+    return AsyncMock()
+
+
+@pytest.mark.asyncio
+async def test_socket_poller_with_data(socket, mocker):
+    socket_data = b"test"
+    socket.recv.return_value = socket_data
+
+    mock_poller: MagicMock = mocker.patch("control_backend.agents.vad_agent.zmq.Poller")
+    mock_poller.return_value.poll.return_value = [(socket, zmq.POLLIN)]
+
+    poller = SocketPoller(socket)
+    # Calling `poll` twice to be able to check that the poller is reused
+    await poller.poll()
+    data = await poller.poll()
+
+    assert data == socket_data
+
+    # Ensure that the poller was reused
+    mock_poller.assert_called_once_with()
+    mock_poller.return_value.register.assert_called_once_with(socket, zmq.POLLIN)
+
+    assert socket.recv.call_count == 2
+
+
+@pytest.mark.asyncio
+async def test_socket_poller_no_data(socket, mocker):
+    mock_poller: MagicMock = mocker.patch("control_backend.agents.vad_agent.zmq.Poller")
+    mock_poller.return_value.poll.return_value = []
+
+    poller = SocketPoller(socket)
+    data = await poller.poll()
+
+    assert data is None
+
+    socket.recv.assert_not_called()

test/unit/agents/test_vad_streaming.py

@@ -21,11 +21,13 @@ def streaming(audio_in_socket, audio_out_socket):
     return Streaming(audio_in_socket, audio_out_socket)
 
 
-@pytest.mark.asyncio
-async def test_voice_activity_detected(audio_in_socket, audio_out_socket, streaming):
-    # After three chunks of audio with speech probability of 1.0, then four chunks of audio with
-    #  speech probability of 0.0, it should send a message over the audio out socket
-    probabilities = [1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0]
+async def simulate_streaming_with_probabilities(streaming, probabilities: list[float]):
+    """
+    Simulates a streaming scenario with given VAD model probabilities for testing purposes.
+
+    :param streaming: The streaming component to be tested.
+    :param probabilities: A list of probabilities representing the outputs of the VAD model.
+    """
     model_item = MagicMock()
     model_item.item.side_effect = probabilities
     streaming.model = MagicMock()
@@ -38,8 +40,53 @@ async def test_voice_activity_detected(audio_in_socket, audio_out_socket, stream
     for _ in probabilities:
         await streaming.run()
 
+
+@pytest.mark.asyncio
+async def test_voice_activity_detected(audio_in_socket, audio_out_socket, streaming):
+    """
+    Test a scenario where there is voice activity detected between silences.
+    :return:
+    """
+    speech_chunk_count = 5
+    probabilities = [0.0]*5 + [1.0]*speech_chunk_count + [0.0]*5
+    await simulate_streaming_with_probabilities(streaming, probabilities)
+
     audio_out_socket.send.assert_called_once()
     data = audio_out_socket.send.call_args[0][0]
     assert isinstance(data, bytes)
-    # each sample has 512 frames of 4 bytes, expecting 5 chunks (3 with speech, 2 as padding)
-    assert len(data) == 512*4*5
+    # each sample has 512 frames of 4 bytes, expecting 7 chunks (5 with speech, 2 as padding)
+    assert len(data) == 512*4*(speech_chunk_count+2)
+
+
+@pytest.mark.asyncio
+async def test_voice_activity_short_pause(audio_in_socket, audio_out_socket, streaming):
+    """
+    Test a scenario where there is a short pause between speech, checking whether it ignores the
+    short pause.
+    """
+    speech_chunk_count = 5
+    probabilities = [0.0]*5 + [1.0]*speech_chunk_count + [0.0] + [1.0]*speech_chunk_count + [0.0]*5
+    await simulate_streaming_with_probabilities(streaming, probabilities)
+
+    audio_out_socket.send.assert_called_once()
+    data = audio_out_socket.send.call_args[0][0]
+    assert isinstance(data, bytes)
+    # Expecting 13 chunks (2*5 with speech, 1 pause between, 2 as padding)
+    assert len(data) == 512*4*(speech_chunk_count*2+1+2)
+
+
+@pytest.mark.asyncio
+async def test_no_data(audio_in_socket, audio_out_socket, streaming):
+    """
+    Test a scenario where there is no data received. This should not cause errors.
+    """
+    audio_in_poller = AsyncMock()
+    audio_in_poller.poll.return_value = None
+    streaming.audio_in_poller = audio_in_poller
+
+    assert streaming.i_since_data == 0
+
+    await streaming.run()
+
+    audio_out_socket.send.assert_not_called()
+    assert streaming.i_since_data == 1