test: complete VAD unit and integration tests

Including an integration test with real voice audio.

ref: N25B-213

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