diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/README.md b/jetson/ros2_ws/src/saltybot_audio_pipeline/README.md
new file mode 100644
index 0000000..9074238
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/README.md
@@ -0,0 +1,39 @@
+# Audio Pipeline (Issue #503)
+
+Comprehensive audio pipeline for Salty Bot with full voice interaction support.
+
+## Features
+
+- **Hardware**: Jabra SPEAK 810 USB audio device integration
+- **Wake Word**: openwakeword "Hey Salty" detection
+- **STT**: whisper.cpp running on Jetson GPU (small/base/medium/large models)
+- **TTS**: Piper synthesis with voice switching
+- **State Machine**: listening → processing → speaking
+- **MQTT**: Real-time status reporting
+- **Metrics**: Latency tracking and performance monitoring
+
+## ROS2 Topics
+
+Published:
+- `/saltybot/speech/transcribed_text` (String): Final STT output
+- `/saltybot/audio/state` (String): Current audio state
+- `/saltybot/audio/status` (String): JSON metrics with latencies
+
+## MQTT Topics
+
+- `saltybot/audio/state`: Current state
+- `saltybot/audio/status`: Complete status JSON
+
+## Launch
+
+```bash
+ros2 launch saltybot_audio_pipeline audio_pipeline.launch.py
+```
+
+## Configuration
+
+See `config/audio_pipeline_params.yaml` for tuning:
+- `device_name`: Jabra device
+- `wake_word_threshold`: 0.5 (0.0-1.0)
+- `whisper_model`: small/base/medium/large
+- `mqtt_enabled`: true/false
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/config/audio_pipeline_params.yaml b/jetson/ros2_ws/src/saltybot_audio_pipeline/config/audio_pipeline_params.yaml
new file mode 100644
index 0000000..ad72268
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/config/audio_pipeline_params.yaml
@@ -0,0 +1,18 @@
+audio_pipeline_node:
+  ros__parameters:
+    device_name: "Jabra SPEAK 810"
+    audio_device_index: -1
+    sample_rate: 16000
+    chunk_size: 512
+    wake_word_model: "hey_salty"
+    wake_word_threshold: 0.5
+    wake_word_timeout_s: 8.0
+    whisper_model: "small"
+    whisper_compute_type: "float16"
+    whisper_language: ""
+    tts_voice_path: "/models/piper/en_US-lessac-medium.onnx"
+    tts_sample_rate: 22050
+    mqtt_enabled: true
+    mqtt_broker: "localhost"
+    mqtt_port: 1883
+    mqtt_base_topic: "saltybot/audio"
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/launch/audio_pipeline.launch.py b/jetson/ros2_ws/src/saltybot_audio_pipeline/launch/audio_pipeline.launch.py
new file mode 100644
index 0000000..5dd06b1
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/launch/audio_pipeline.launch.py
@@ -0,0 +1,19 @@
+#!/usr/bin/env python3
+from launch import LaunchDescription
+from launch_ros.actions import Node
+from ament_index_python.packages import get_package_share_directory
+import os
+
+def generate_launch_description():
+    pkg_dir = get_package_share_directory("saltybot_audio_pipeline")
+    config_path = os.path.join(pkg_dir, "config", "audio_pipeline_params.yaml")
+    return LaunchDescription([
+        Node(
+            package="saltybot_audio_pipeline",
+            executable="audio_pipeline_node",
+            name="audio_pipeline_node",
+            parameters=[config_path],
+            output="screen",
+            emulate_tty=True,
+        ),
+    ])
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/package.xml b/jetson/ros2_ws/src/saltybot_audio_pipeline/package.xml
new file mode 100644
index 0000000..b5bee53
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/package.xml
@@ -0,0 +1,12 @@
+<?xml version="1.0"?>
+<package format="3">
+  <name>saltybot_audio_pipeline</name>
+  <version>1.0.0</version>
+  <description>Full audio pipeline: Jabra SPEAK 810, wake word, STT, TTS with MQTT (Issue #503)</description>
+  <maintainer email="seb@saltylab.ai">Salty Lab</maintainer>
+  <license>Apache-2.0</license>
+  <buildtool_depend>ament_python</buildtool_depend>
+  <depend>rclpy</depend>
+  <depend>std_msgs</depend>
+  <test_depend>pytest</test_depend>
+</package>
\ No newline at end of file
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/resource/saltybot_audio_pipeline b/jetson/ros2_ws/src/saltybot_audio_pipeline/resource/saltybot_audio_pipeline
new file mode 100644
index 0000000..e69de29
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/__init__.py b/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/__init__.py
new file mode 100644
index 0000000..f6ffb99
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/__init__.py
@@ -0,0 +1,2 @@
+"""Audio pipeline for Salty Bot."""
+__version__ = "1.0.0"
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/audio_pipeline_node.py b/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/audio_pipeline_node.py
new file mode 100644
index 0000000..0cc9cdc
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/audio_pipeline_node.py
@@ -0,0 +1,380 @@
+#!/usr/bin/env python3
+"""audio_pipeline_node.py — Full audio pipeline with Jabra SPEAK 810 I/O (Issue #503)."""
+
+from __future__ import annotations
+import json, os, threading, time
+from enum import Enum
+from dataclasses import dataclass, asdict
+from typing import Optional
+
+import rclpy
+from rclpy.node import Node
+from rclpy.qos import QoSProfile
+from std_msgs.msg import String
+
+from .audio_utils import EnergyVAD, UtteranceSegmenter, AudioBuffer, pcm16_to_float32, float32_to_pcm16, resample_audio
+
+try:
+    import paho.mqtt.client as mqtt
+    _MQTT_AVAILABLE = True
+except ImportError:
+    _MQTT_AVAILABLE = False
+
+
+class AudioState(Enum):
+    IDLE = "idle"
+    LISTENING = "listening"
+    WAKE_WORD_DETECTED = "wake_detected"
+    PROCESSING = "processing"
+    SPEAKING = "speaking"
+    ERROR = "error"
+
+
+@dataclass
+class AudioMetrics:
+    wake_to_stt_ms: float = 0.0
+    stt_processing_ms: float = 0.0
+    tts_synthesis_ms: float = 0.0
+    total_latency_ms: float = 0.0
+    transcribed_text: str = ""
+    speaker_id: str = "unknown"
+    error_msg: str = ""
+
+
+class MqttClient:
+    def __init__(self, broker: str, port: int, base_topic: str):
+        self.broker = broker
+        self.port = port
+        self.base_topic = base_topic
+        self._client = None
+        self._connected = False
+        if _MQTT_AVAILABLE:
+            try:
+                self._client = mqtt.Client(client_id=f"saltybot-audio-{int(time.time())}")
+                self._client.on_connect = lambda c, u, f, rc: setattr(self, '_connected', rc == 0)
+                self._client.on_disconnect = lambda c, u, rc: setattr(self, '_connected', False)
+                self._client.connect_async(broker, port, keepalive=60)
+                self._client.loop_start()
+            except Exception as e:
+                print(f"MQTT init failed: {e}")
+
+    def publish(self, topic: str, payload: str) -> bool:
+        if not self._client or not self._connected:
+            return False
+        try:
+            self._client.publish(topic, payload, qos=0)
+            return True
+        except Exception:
+            return False
+
+    def disconnect(self) -> None:
+        if self._client:
+            self._client.loop_stop()
+            self._client.disconnect()
+
+
+class JabraAudioDevice:
+    def __init__(self, device_name: str = "Jabra SPEAK 810", device_idx: int = -1):
+        self.device_name = device_name
+        self.device_idx = device_idx
+        self._pa = None
+        self._input_stream = None
+        self._output_stream = None
+        self._is_open = False
+
+    def open(self, sample_rate: int = 16000, chunk_size: int = 512) -> bool:
+        try:
+            import pyaudio
+            self._pa = pyaudio.PyAudio()
+            if self.device_idx < 0:
+                self.device_idx = self._find_device_index() or None
+            self._input_stream = self._pa.open(format=pyaudio.paInt16, channels=1, rate=sample_rate,
+                input=True, input_device_index=self.device_idx, frames_per_buffer=chunk_size, start=False)
+            self._output_stream = self._pa.open(format=pyaudio.paInt16, channels=1, rate=sample_rate,
+                output=True, output_device_index=self.device_idx, frames_per_buffer=chunk_size, start=False)
+            self._is_open = True
+            return True
+        except Exception as e:
+            print(f"Failed to open Jabra device: {e}")
+            return False
+
+    def _find_device_index(self) -> int:
+        try:
+            import pyaudio
+            pa = pyaudio.PyAudio()
+            for i in range(pa.get_device_count()):
+                info = pa.get_device_info_by_index(i)
+                if "jabra" in info["name"].lower() or "speak" in info["name"].lower():
+                    return i
+        except Exception:
+            pass
+        return -1
+
+    def read_chunk(self, chunk_size: int = 512) -> Optional[bytes]:
+        if not self._is_open or not self._input_stream:
+            return None
+        try:
+            return self._input_stream.read(chunk_size, exception_on_overflow=False)
+        except Exception:
+            return None
+
+    def write_chunk(self, pcm_data: bytes) -> bool:
+        if not self._is_open or not self._output_stream:
+            return False
+        try:
+            self._output_stream.write(pcm_data)
+            return True
+        except Exception:
+            return False
+
+    def close(self) -> None:
+        if self._input_stream:
+            self._input_stream.stop_stream()
+            self._input_stream.close()
+        if self._output_stream:
+            self._output_stream.stop_stream()
+            self._output_stream.close()
+        if self._pa:
+            self._pa.terminate()
+        self._is_open = False
+
+
+class AudioPipelineNode(Node):
+    def __init__(self) -> None:
+        super().__init__("audio_pipeline_node")
+        for param, default in [
+            ("device_name", "Jabra SPEAK 810"),
+            ("audio_device_index", -1),
+            ("sample_rate", 16000),
+            ("chunk_size", 512),
+            ("wake_word_model", "hey_salty"),
+            ("wake_word_threshold", 0.5),
+            ("wake_word_timeout_s", 8.0),
+            ("whisper_model", "small"),
+            ("whisper_compute_type", "float16"),
+            ("whisper_language", ""),
+            ("tts_voice_path", "/models/piper/en_US-lessac-medium.onnx"),
+            ("tts_sample_rate", 22050),
+            ("mqtt_enabled", True),
+            ("mqtt_broker", "localhost"),
+            ("mqtt_port", 1883),
+            ("mqtt_base_topic", "saltybot/audio"),
+        ]:
+            self.declare_parameter(param, default)
+
+        device_name = self.get_parameter("device_name").value
+        device_idx = self.get_parameter("audio_device_index").value
+        self._sample_rate = self.get_parameter("sample_rate").value
+        self._chunk_size = self.get_parameter("chunk_size").value
+        self._ww_model = self.get_parameter("wake_word_model").value
+        self._ww_thresh = self.get_parameter("wake_word_threshold").value
+        self._whisper_model = self.get_parameter("whisper_model").value
+        self._compute_type = self.get_parameter("whisper_compute_type").value
+        self._whisper_lang = self.get_parameter("whisper_language").value or None
+        self._tts_voice_path = self.get_parameter("tts_voice_path").value
+        self._tts_rate = self.get_parameter("tts_sample_rate").value
+        mqtt_enabled = self.get_parameter("mqtt_enabled").value
+        mqtt_broker = self.get_parameter("mqtt_broker").value
+        mqtt_port = self.get_parameter("mqtt_port").value
+        mqtt_topic = self.get_parameter("mqtt_base_topic").value
+
+        qos = QoSProfile(depth=10)
+        self._text_pub = self.create_publisher(String, "/saltybot/speech/transcribed_text", qos)
+        self._state_pub = self.create_publisher(String, "/saltybot/audio/state", qos)
+        self._status_pub = self.create_publisher(String, "/saltybot/audio/status", qos)
+
+        self._state = AudioState.IDLE
+        self._state_lock = threading.Lock()
+        self._metrics = AudioMetrics()
+        self._running = False
+
+        self._jabra = JabraAudioDevice(device_name, device_idx)
+        self._oww = None
+        self._whisper = None
+        self._tts_voice = None
+
+        self._mqtt = None
+        if mqtt_enabled and _MQTT_AVAILABLE:
+            try:
+                self._mqtt = MqttClient(mqtt_broker, mqtt_port, mqtt_topic)
+                self.get_logger().info(f"MQTT enabled: {mqtt_broker}:{mqtt_port}/{mqtt_topic}")
+            except Exception as e:
+                self.get_logger().warn(f"MQTT init failed: {e}")
+
+        self._vad = EnergyVAD(threshold_db=-35.0)
+        self._segmenter = UtteranceSegmenter(self._vad, sample_rate=self._sample_rate)
+        self._audio_buffer = AudioBuffer(capacity_s=30.0, sample_rate=self._sample_rate)
+
+        threading.Thread(target=self._init_pipeline, daemon=True).start()
+
+    def _init_pipeline(self) -> None:
+        self.get_logger().info("Initializing audio pipeline...")
+        t0 = time.time()
+
+        if not self._jabra.open(self._sample_rate, self._chunk_size):
+            self._set_state(AudioState.ERROR)
+            self._metrics.error_msg = "Failed to open Jabra device"
+            return
+
+        try:
+            from openwakeword.model import Model as OWWModel
+            self._oww = OWWModel(wakeword_models=[self._ww_model])
+            self.get_logger().info(f"openwakeword '{self._ww_model}' loaded")
+        except Exception as e:
+            self.get_logger().warn(f"openwakeword failed: {e}")
+
+        try:
+            from faster_whisper import WhisperModel
+            self._whisper = WhisperModel(self._whisper_model, device="cuda",
+                compute_type=self._compute_type, download_root="/models")
+            self.get_logger().info(f"Whisper '{self._whisper_model}' loaded")
+        except Exception as e:
+            self.get_logger().error(f"Whisper failed: {e}")
+            self._set_state(AudioState.ERROR)
+            self._metrics.error_msg = f"Whisper init: {e}"
+            return
+
+        try:
+            from piper import PiperVoice
+            self._tts_voice = PiperVoice.load(self._tts_voice_path)
+            self.get_logger().info("Piper TTS loaded")
+        except Exception as e:
+            self.get_logger().warn(f"Piper TTS failed: {e}")
+
+        self.get_logger().info(f"Audio pipeline ready ({time.time()-t0:.1f}s)")
+        self._set_state(AudioState.LISTENING)
+        self._publish_status()
+
+        threading.Thread(target=self._audio_loop, daemon=True).start()
+
+    def _audio_loop(self) -> None:
+        self._running = True
+        import numpy as np
+        while self._running and self._state != AudioState.ERROR:
+            raw_chunk = self._jabra.read_chunk(self._chunk_size)
+            if raw_chunk is None:
+                continue
+            samples = pcm16_to_float32(raw_chunk)
+            self._audio_buffer.push(samples)
+
+            if self._state == AudioState.LISTENING and self._oww is not None:
+                try:
+                    preds = self._oww.predict(samples)
+                    score = preds.get(self._ww_model, 0.0)
+                    if isinstance(score, (list, tuple)):
+                        score = score[-1]
+                    if score >= self._ww_thresh:
+                        self.get_logger().info(f"Wake word detected (score={score:.2f})")
+                        self._metrics.wake_to_stt_ms = 0.0
+                        self._set_state(AudioState.WAKE_WORD_DETECTED)
+                        self._segmenter.reset()
+                        self._audio_buffer.clear()
+                except Exception as e:
+                    self.get_logger().debug(f"Wake word error: {e}")
+
+            if self._state == AudioState.WAKE_WORD_DETECTED:
+                completed = self._segmenter.push(samples)
+                for utt_samples, duration in completed:
+                    threading.Thread(target=self._process_utterance,
+                        args=(utt_samples, duration), daemon=True).start()
+
+    def _process_utterance(self, audio_samples: list, duration: float) -> None:
+        if self._whisper is None:
+            self._set_state(AudioState.LISTENING)
+            return
+        self._set_state(AudioState.PROCESSING)
+        t0 = time.time()
+        try:
+            import numpy as np
+            audio_np = np.array(audio_samples, dtype=np.float32) if isinstance(audio_samples, list) else audio_samples.astype(np.float32)
+            segments_gen, info = self._whisper.transcribe(audio_np, language=self._whisper_lang, beam_size=3, vad_filter=False)
+            text = " ".join([seg.text.strip() for seg in segments_gen]).strip()
+            if text:
+                stt_time = (time.time() - t0) * 1000
+                self._metrics.stt_processing_ms = stt_time
+                self._metrics.transcribed_text = text
+                self._metrics.total_latency_ms = stt_time
+                msg = String()
+                msg.data = text
+                self._text_pub.publish(msg)
+                self.get_logger().info(f"STT [{duration:.1f}s, {stt_time:.0f}ms]: '{text}'")
+                self._process_tts(text)
+            else:
+                self._set_state(AudioState.LISTENING)
+        except Exception as e:
+            self.get_logger().error(f"STT error: {e}")
+            self._metrics.error_msg = str(e)
+            self._set_state(AudioState.LISTENING)
+
+    def _process_tts(self, text: str) -> None:
+        if self._tts_voice is None:
+            self._set_state(AudioState.LISTENING)
+            return
+        self._set_state(AudioState.SPEAKING)
+        t0 = time.time()
+        try:
+            pcm_data = b"".join(self._tts_voice.synthesize_stream_raw(text))
+            self._metrics.tts_synthesis_ms = (time.time() - t0) * 1000
+            if self._tts_rate != self._sample_rate:
+                import numpy as np
+                samples = np.frombuffer(pcm_data, dtype=np.int16).astype(np.float32) / 32768.0
+                pcm_data = float32_to_pcm16(resample_audio(samples, self._tts_rate, self._sample_rate))
+            self._jabra.write_chunk(pcm_data)
+            self.get_logger().info(f"TTS: played {len(pcm_data)} bytes")
+        except Exception as e:
+            self.get_logger().error(f"TTS error: {e}")
+            self._metrics.error_msg = str(e)
+        finally:
+            self._set_state(AudioState.LISTENING)
+            self._publish_status()
+
+    def _set_state(self, new_state: AudioState) -> None:
+        with self._state_lock:
+            if self._state != new_state:
+                self._state = new_state
+                self.get_logger().info(f"Audio state: {new_state.value}")
+                msg = String()
+                msg.data = new_state.value
+                self._state_pub.publish(msg)
+                if self._mqtt:
+                    try:
+                        self._mqtt.publish(f"{self._mqtt.base_topic}/state", new_state.value)
+                    except Exception as e:
+                        self.get_logger().debug(f"MQTT publish failed: {e}")
+
+    def _publish_status(self) -> None:
+        status = {"state": self._state.value, "metrics": asdict(self._metrics), "timestamp": time.time()}
+        msg = String()
+        msg.data = json.dumps(status)
+        self._status_pub.publish(msg)
+        if self._mqtt:
+            try:
+                self._mqtt.publish(f"{self._mqtt.base_topic}/status", msg.data)
+            except Exception as e:
+                self.get_logger().debug(f"MQTT publish failed: {e}")
+
+    def destroy_node(self) -> None:
+        self._running = False
+        self._jabra.close()
+        if self._mqtt:
+            try:
+                self._mqtt.disconnect()
+            except Exception:
+                pass
+        super().destroy_node()
+
+
+def main(args=None) -> None:
+    rclpy.init(args=args)
+    node = AudioPipelineNode()
+    try:
+        rclpy.spin(node)
+    except KeyboardInterrupt:
+        pass
+    finally:
+        node.destroy_node()
+        rclpy.shutdown()
+
+
+if __name__ == "__main__":
+    main()
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/audio_utils.py b/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/audio_utils.py
new file mode 100644
index 0000000..b00f6ba
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/saltybot_audio_pipeline/audio_utils.py
@@ -0,0 +1,133 @@
+"""Audio utilities for processing and buffering."""
+
+from __future__ import annotations
+from typing import Optional, Tuple, List
+import threading, time
+from collections import deque
+from dataclasses import dataclass
+import numpy as np
+
+
+@dataclass
+class AudioChunk:
+    samples: np.ndarray
+    timestamp: float
+    rms_db: float
+
+
+class EnergyVAD:
+    """Energy-based Voice Activity Detection."""
+    def __init__(self, threshold_db: float = -35.0):
+        self.threshold_db = threshold_db
+
+    def is_speech(self, samples: np.ndarray) -> bool:
+        rms = np.sqrt(np.mean(samples ** 2))
+        db = 20 * np.log10(rms + 1e-10)
+        return db > self.threshold_db
+
+    def rms_db(self, samples: np.ndarray) -> float:
+        rms = np.sqrt(np.mean(samples ** 2))
+        return 20 * np.log10(rms + 1e-10)
+
+
+class UtteranceSegmenter:
+    """Buffer and segment audio utterances based on energy VAD."""
+    def __init__(self, vad: Optional[EnergyVAD] = None, silence_duration_s: float = 0.5,
+                 min_duration_s: float = 0.3, sample_rate: int = 16000):
+        self.vad = vad or EnergyVAD()
+        self.silence_duration_s = silence_duration_s
+        self.min_duration_s = min_duration_s
+        self.sample_rate = sample_rate
+        self._buffer = deque()
+        self._last_speech_time = 0.0
+        self._speech_started = False
+        self._lock = threading.Lock()
+
+    def push(self, samples: np.ndarray) -> List[Tuple[List[float], float]]:
+        completed = []
+        with self._lock:
+            now = time.time()
+            is_speech = self.vad.is_speech(samples)
+            if is_speech:
+                self._last_speech_time = now
+                self._speech_started = True
+                self._buffer.append(samples)
+            else:
+                self._buffer.append(samples)
+                if self._speech_started and now - self._last_speech_time > self.silence_duration_s:
+                    utt_samples = self._extract_buffer()
+                    duration = len(utt_samples) / self.sample_rate
+                    if duration >= self.min_duration_s:
+                        completed.append((utt_samples, duration))
+                    self._speech_started = False
+                    self._buffer.clear()
+        return completed
+
+    def _extract_buffer(self) -> List[float]:
+        if not self._buffer:
+            return []
+        flat = []
+        for s in self._buffer:
+            flat.extend(s.tolist() if isinstance(s, np.ndarray) else s)
+        return flat
+
+    def reset(self) -> None:
+        with self._lock:
+            self._buffer.clear()
+            self._speech_started = False
+
+
+class AudioBuffer:
+    """Thread-safe circular audio buffer."""
+    def __init__(self, capacity_s: float = 5.0, sample_rate: int = 16000):
+        self.capacity = int(capacity_s * sample_rate)
+        self.sample_rate = sample_rate
+        self._buffer = deque(maxlen=self.capacity)
+        self._lock = threading.Lock()
+
+    def push(self, samples: np.ndarray) -> None:
+        with self._lock:
+            self._buffer.extend(samples.tolist() if isinstance(samples, np.ndarray) else samples)
+
+    def extract(self, duration_s: Optional[float] = None) -> np.ndarray:
+        with self._lock:
+            samples = list(self._buffer)
+            if duration_s is not None:
+                num_samples = int(duration_s * self.sample_rate)
+                samples = samples[-num_samples:]
+            return np.array(samples, dtype=np.float32)
+
+    def clear(self) -> None:
+        with self._lock:
+            self._buffer.clear()
+
+    def size(self) -> int:
+        with self._lock:
+            return len(self._buffer)
+
+
+def pcm16_to_float32(pcm_bytes: bytes) -> np.ndarray:
+    samples = np.frombuffer(pcm_bytes, dtype=np.int16)
+    return samples.astype(np.float32) / 32768.0
+
+
+def float32_to_pcm16(samples: np.ndarray) -> bytes:
+    if isinstance(samples, list):
+        samples = np.array(samples, dtype=np.float32)
+    clipped = np.clip(samples, -1.0, 1.0)
+    pcm = (clipped * 32767).astype(np.int16)
+    return pcm.tobytes()
+
+
+def resample_audio(samples: np.ndarray, orig_rate: int, target_rate: int) -> np.ndarray:
+    if orig_rate == target_rate:
+        return samples
+    from scipy import signal
+    num_samples = int(len(samples) * target_rate / orig_rate)
+    resampled = signal.resample(samples, num_samples)
+    return resampled.astype(np.float32)
+
+
+def calculate_rms_db(samples: np.ndarray) -> float:
+    rms = np.sqrt(np.mean(samples ** 2))
+    return 20 * np.log10(rms + 1e-10)
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/setup.cfg b/jetson/ros2_ws/src/saltybot_audio_pipeline/setup.cfg
new file mode 100644
index 0000000..633b438
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/setup.cfg
@@ -0,0 +1,2 @@
+[develop]
+script_dir=$base/lib/saltybot_audio_pipeline/scripts
\ No newline at end of file
diff --git a/jetson/ros2_ws/src/saltybot_audio_pipeline/setup.py b/jetson/ros2_ws/src/saltybot_audio_pipeline/setup.py
new file mode 100644
index 0000000..214aa20
--- /dev/null
+++ b/jetson/ros2_ws/src/saltybot_audio_pipeline/setup.py
@@ -0,0 +1,21 @@
+from setuptools import setup
+package_name = 'saltybot_audio_pipeline'
+setup(
+    name=package_name,
+    version='1.0.0',
+    packages=[package_name],
+    data_files=[
+        ('share/ament_index/resource_index/packages', ['resource/' + package_name]),
+        ('share/' + package_name, ['package.xml']),
+        ('share/' + package_name + '/launch', ['launch/audio_pipeline.launch.py']),
+        ('share/' + package_name + '/config', ['config/audio_pipeline_params.yaml']),
+    ],
+    install_requires=['setuptools'],
+    zip_safe=True,
+    author='Salty Lab',
+    entry_points={
+        'console_scripts': [
+            'audio_pipeline_node = saltybot_audio_pipeline.audio_pipeline_node:main',
+        ],
+    },
+)
\ No newline at end of file