diff --git a/tests/test_audio_buffer_processor.py b/tests/test_audio_buffer_processor.py
new file mode 100644
index 000000000..24ae42265
--- /dev/null
+++ b/tests/test_audio_buffer_processor.py
@@ -0,0 +1,111 @@
+#
+# Copyright (c) 2024-2025 Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+import asyncio
+import struct
+import unittest
+
+from pipecat.frames.frames import InputAudioRawFrame, OutputAudioRawFrame, StartFrame
+from pipecat.processors.audio.audio_buffer_processor import AudioBufferProcessor
+
+
+class _PassthroughResampler:
+    async def resample(self, audio: bytes, in_rate: int, out_rate: int) -> bytes:  # pragma: no cover - trivial
+        return audio
+
+
+class TestAudioBufferProcessor(unittest.IsolatedAsyncioTestCase):
+    async def asyncSetUp(self):
+        self.processor = AudioBufferProcessor(sample_rate=16000, num_channels=2, buffer_size=4)
+        self.processor._input_resampler = _PassthroughResampler()
+        self.processor._output_resampler = _PassthroughResampler()
+        self.processor._update_sample_rate(StartFrame(audio_out_sample_rate=16000))
+        await self.processor.start_recording()
+
+    async def asyncTearDown(self):
+        if getattr(self.processor, "_recording", False):
+            await self.processor.stop_recording()
+        await self.processor.cleanup()
+
+    async def test_flush_user_audio_pads_bot_track(self):
+        user_audio = struct.pack("<hh", 1000, -1000)
+        audio_event = asyncio.Event()
+        track_event = asyncio.Event()
+        captured = {}
+
+        async def on_audio_data(_, audio: bytes, sample_rate: int, num_channels: int):
+            captured["merged"] = (audio, sample_rate, num_channels)
+            audio_event.set()
+
+        async def on_track_audio_data(_, user: bytes, bot: bytes, sample_rate: int, num_channels: int):
+            captured["tracks"] = (user, bot, sample_rate, num_channels)
+            track_event.set()
+
+        self.processor.add_event_handler("on_audio_data", on_audio_data)
+        self.processor.add_event_handler("on_track_audio_data", on_track_audio_data)
+
+        frame = InputAudioRawFrame(audio=user_audio, sample_rate=16000, num_channels=1)
+        await self.processor._process_recording(frame)
+
+        await asyncio.wait_for(audio_event.wait(), timeout=1)
+        await asyncio.wait_for(track_event.wait(), timeout=1)
+
+        merged_audio, merged_sr, merged_channels = captured["merged"]
+        user_track, bot_track, track_sr, track_channels = captured["tracks"]
+
+        self.assertEqual(merged_sr, 16000)
+        self.assertEqual(merged_channels, 2)
+        self.assertEqual(track_sr, 16000)
+        self.assertEqual(track_channels, 2)
+        self.assertEqual(user_track, user_audio)
+        self.assertEqual(bot_track, b"\x00" * len(user_audio))
+        self.assertEqual(len(merged_audio), len(user_audio) * 2)
+        self.assertEqual(merged_audio[0:2], user_audio[0:2])
+        self.assertEqual(merged_audio[2:4], b"\x00\x00")
+        self.assertEqual(merged_audio[4:6], user_audio[2:4])
+        self.assertEqual(merged_audio[6:8], b"\x00\x00")
+        self.assertEqual(len(self.processor._user_audio_buffer), 0)
+        self.assertEqual(len(self.processor._bot_audio_buffer), 0)
+
+    async def test_flush_bot_audio_pads_user_track(self):
+        bot_audio = struct.pack("<hh", -800, 400)
+        audio_event = asyncio.Event()
+        track_event = asyncio.Event()
+        captured = {}
+
+        async def on_audio_data(_, audio: bytes, sample_rate: int, num_channels: int):
+            captured["merged"] = (audio, sample_rate, num_channels)
+            audio_event.set()
+
+        async def on_track_audio_data(_, user: bytes, bot: bytes, sample_rate: int, num_channels: int):
+            captured["tracks"] = (user, bot, sample_rate, num_channels)
+            track_event.set()
+
+        self.processor.add_event_handler("on_audio_data", on_audio_data)
+        self.processor.add_event_handler("on_track_audio_data", on_track_audio_data)
+
+        frame = OutputAudioRawFrame(audio=bot_audio, sample_rate=16000, num_channels=1)
+        await self.processor._process_recording(frame)
+
+        await asyncio.wait_for(audio_event.wait(), timeout=1)
+        await asyncio.wait_for(track_event.wait(), timeout=1)
+
+        merged_audio, merged_sr, merged_channels = captured["merged"]
+        user_track, bot_track, track_sr, track_channels = captured["tracks"]
+
+        self.assertEqual(merged_sr, 16000)
+        self.assertEqual(merged_channels, 2)
+        self.assertEqual(track_sr, 16000)
+        self.assertEqual(track_channels, 2)
+        self.assertEqual(user_track, b"\x00" * len(bot_audio))
+        self.assertEqual(bot_track, bot_audio)
+        self.assertEqual(len(merged_audio), len(bot_audio) * 2)
+        self.assertEqual(merged_audio[0:2], b"\x00\x00")
+        self.assertEqual(merged_audio[2:4], bot_audio[0:2])
+        self.assertEqual(merged_audio[4:6], b"\x00\x00")
+        self.assertEqual(merged_audio[6:8], bot_audio[2:4])
+        self.assertEqual(len(self.processor._user_audio_buffer), 0)
+        self.assertEqual(len(self.processor._bot_audio_buffer), 0)