diff --git a/src/fenrirscreenreader/speechDriver/hardwareSerialDriver.py b/src/fenrirscreenreader/speechDriver/hardwareSerialDriver.py
index e006b865..a32cdd8b 100644
--- a/src/fenrirscreenreader/speechDriver/hardwareSerialDriver.py
+++ b/src/fenrirscreenreader/speechDriver/hardwareSerialDriver.py
@@ -6,6 +6,7 @@
 
 import glob
 import os
+import select
 import termios
 import threading
 import tty
@@ -28,6 +29,8 @@ class SpeakQueue(Queue):
 class hardware_serial_driver(speech_driver):
     cancel_command = b""
     default_baud_rate = 9600
+    hardware_probe_command = b""
+    hardware_probe_timeout = 0.2
 
     def __init__(self):
         speech_driver.__init__(self)
@@ -43,14 +46,22 @@ class hardware_serial_driver(speech_driver):
         self.env = environment
         self._is_initialized = False
         settings_manager = self.env["runtime"]["SettingsManager"]
-        self.device = settings_manager.get_setting(
-            "speech", "hardware_device"
+        self.device = self._clean_device_setting(
+            settings_manager.get_setting("speech", "hardware_device")
         )
         self.baud_rate = settings_manager.get_setting_as_int(
             "speech", "hardware_baud_rate"
         )
+        self._debug(
+            "Hardware speech initialize: "
+            f"requested_device={self.device}, baud_rate={self.baud_rate}",
+            debug.DebugLevel.INFO,
+            on_any_level=True,
+        )
         self._open_serial_port()
         self._is_initialized = self.serial_port is not None
+        if not self._is_initialized:
+            raise RuntimeError("hardware speech device is not available")
         if self._is_initialized:
             self._stop_worker = False
             self.worker_thread = threading.Thread(
@@ -58,6 +69,12 @@ class hardware_serial_driver(speech_driver):
             )
             self.worker_thread.start()
 
+    def _clean_device_setting(self, device):
+        if not isinstance(device, str):
+            return "auto"
+        device = device.split("#", 1)[0].split(";", 1)[0].strip()
+        return device or "auto"
+
     def shutdown(self):
         if not self._is_initialized:
             return
@@ -76,6 +93,12 @@ class hardware_serial_driver(speech_driver):
             self.cancel()
         if not isinstance(text, str) or text == "":
             return
+        self._debug(
+            "Hardware speech queued text: "
+            f"{len(text)} chars, queue_size={self.text_queue.qsize()}",
+            debug.DebugLevel.INFO,
+            on_any_level=True,
+        )
         self.text_queue.put(text)
 
     def cancel(self):
@@ -83,7 +106,7 @@ class hardware_serial_driver(speech_driver):
             return
         self.clear_buffer()
         if self.cancel_command:
-            self._write_bytes(self.cancel_command)
+            self._write_bytes(self.cancel_command, "cancel")
 
     def clear_buffer(self):
         if not self._is_initialized:
@@ -95,37 +118,88 @@ class hardware_serial_driver(speech_driver):
             return
         if not isinstance(rate, float):
             return
-        self._write_bytes(self._rate_command(rate))
+        self._write_bytes(self._rate_command(rate), "rate")
 
     def set_pitch(self, pitch):
         if not self._is_initialized:
             return
         if not isinstance(pitch, float):
             return
-        self._write_bytes(self._pitch_command(pitch))
+        self._write_bytes(self._pitch_command(pitch), "pitch")
 
     def set_volume(self, volume):
         if not self._is_initialized:
             return
         if not isinstance(volume, float):
             return
-        self._write_bytes(self._volume_command(volume))
+        self._write_bytes(self._volume_command(volume), "volume")
 
     def _worker(self):
         while not self._stop_worker:
             text = self.text_queue.get()
             if text is None:
                 return
-            self._write_bytes(self._speak_bytes(text))
+            try:
+                data = self._speak_bytes(text)
+                self._debug(
+                    "Hardware speech worker prepared speech bytes: "
+                    f"{len(data)} bytes",
+                    debug.DebugLevel.INFO,
+                    on_any_level=True,
+                )
+                self._write_bytes(data, "speech")
+            except Exception as error:
+                self._debug(
+                    f"Hardware speech worker failed: {error}",
+                    debug.DebugLevel.ERROR,
+                    on_any_level=True,
+                )
 
     def _open_serial_port(self):
-        device = self._resolve_device(self.device)
-        if not device:
+        auto_detect = self.device == "auto"
+        devices = self._resolve_devices(self.device)
+        if not devices:
             self._debug(
                 "Hardware speech device not found",
                 debug.DebugLevel.ERROR,
+                on_any_level=True,
             )
             return
+
+        fallback_device = None
+        fallback_port = None
+        for device in devices:
+            port = self._open_configured_serial_port(device)
+            if port is None:
+                continue
+            if not auto_detect:
+                self._activate_serial_port(
+                    device, port, probe_matched=False
+                )
+                return
+            if self._probe_serial_port(port):
+                if fallback_port is not None:
+                    self._close_port(fallback_port)
+                self._activate_serial_port(device, port, probe_matched=True)
+                return
+            if fallback_port is None:
+                fallback_device = device
+                fallback_port = port
+            else:
+                self._close_port(port)
+
+        if fallback_port is not None:
+            self._debug(
+                "Hardware speech probe did not identify a synth; "
+                f"falling back to {fallback_device}",
+                debug.DebugLevel.WARNING,
+                on_any_level=True,
+            )
+            self._activate_serial_port(
+                fallback_device, fallback_port, probe_matched=False
+            )
+
+    def _open_configured_serial_port(self, device):
         try:
             port = os.open(device, os.O_RDWR | os.O_NOCTTY)
             tty.setraw(port)
@@ -138,51 +212,147 @@ class hardware_serial_driver(speech_driver):
             attrs[6][termios.VTIME] = 0
             attrs[0] &= ~(termios.IXON | termios.IXOFF | termios.IXANY)
             termios.tcsetattr(port, termios.TCSANOW, attrs)
-            self.serial_port = port
-            self.device = device
-        except OSError as error:
+            return port
+        except (OSError, termios.error) as error:
             self._debug(
                 f"Hardware speech device open failed: {device}: {error}",
                 debug.DebugLevel.ERROR,
+                on_any_level=True,
             )
-            self.serial_port = None
+            return None
+
+    def _activate_serial_port(self, device, port, probe_matched=False):
+        self.serial_port = port
+        self.device = device
+        probe_status = "matched" if probe_matched else "not matched"
+        self._debug(
+            "Hardware speech device opened: "
+            f"{device}, baud_rate={self.baud_rate}, probe={probe_status}",
+            debug.DebugLevel.INFO,
+            on_any_level=True,
+        )
+
+    def _probe_serial_port(self, port):
+        if not self.hardware_probe_command:
+            return True
+        try:
+            self._drain_serial_input(port)
+            os.write(port, self.hardware_probe_command)
+            readable, _, _ = select.select(
+                [port], [], [], self.hardware_probe_timeout
+            )
+            if not readable:
+                self._debug(
+                    "Hardware speech probe got no response",
+                    debug.DebugLevel.INFO,
+                    on_any_level=True,
+                )
+                return False
+            response = os.read(port, 256)
+            self._debug(
+                "Hardware speech probe response: "
+                f"{self._format_bytes_preview(response)}",
+                debug.DebugLevel.INFO,
+                on_any_level=True,
+            )
+            return self._is_hardware_probe_response(response)
+        except OSError as error:
+            self._debug(
+                f"Hardware speech probe failed: {error}",
+                debug.DebugLevel.WARNING,
+                on_any_level=True,
+            )
+            return False
+
+    def _drain_serial_input(self, port):
+        while True:
+            readable, _, _ = select.select([port], [], [], 0)
+            if not readable:
+                return
+            if not os.read(port, 256):
+                return
+
+    def _is_hardware_probe_response(self, response):
+        return bool(response)
 
     def _close_serial_port(self):
         with self.lock:
             if self.serial_port is None:
                 return
-            try:
-                os.close(self.serial_port)
-            except OSError as error:
-                self._debug(
-                    f"Hardware speech device close failed: {error}",
-                    debug.DebugLevel.WARNING,
-                )
-            finally:
-                self.serial_port = None
+            self._close_port(self.serial_port)
+            self.serial_port = None
 
-    def _write_bytes(self, data):
+    def _close_port(self, port):
+        if port is None:
+            return
+        try:
+            os.close(port)
+        except OSError as error:
+            self._debug(
+                f"Hardware speech device close failed: {error}",
+                debug.DebugLevel.WARNING,
+            )
+
+    def _write_bytes(self, data, description="data"):
         if not data:
             return
         with self.lock:
             if self.serial_port is None:
                 return
             try:
-                os.write(self.serial_port, data)
+                total_written = 0
+                while total_written < len(data):
+                    bytes_written = os.write(
+                        self.serial_port, data[total_written:]
+                    )
+                    if bytes_written == 0:
+                        raise OSError("serial write returned 0 bytes")
+                    total_written += bytes_written
+                preview = self._format_bytes_preview(data)
+                self._debug(
+                    "Hardware speech wrote "
+                    f"{total_written} {description} bytes: {preview}",
+                    debug.DebugLevel.INFO,
+                    on_any_level=True,
+                )
             except OSError as error:
                 self._debug(
                     f"Hardware speech write failed: {error}",
                     debug.DebugLevel.ERROR,
+                    on_any_level=True,
                 )
 
-    def _resolve_device(self, device):
+    def _resolve_devices(self, device):
         if device and device != "auto":
-            return device
-        for pattern in ("/dev/ttyACM*", "/dev/ttyUSB*"):
+            self._debug(
+                f"Hardware speech using configured device: {device}",
+                debug.DebugLevel.INFO,
+                on_any_level=True,
+            )
+            return [device]
+        devices = []
+        for pattern in ("/dev/ttyACM*", "/dev/ttyUSB*", "/dev/ttyS*"):
             matches = sorted(glob.glob(pattern))
+            self._debug(
+                f"Hardware speech auto scan {pattern}: {matches}",
+                debug.DebugLevel.INFO,
+                on_any_level=True,
+            )
             if matches:
-                return matches[0]
-        return ""
+                if len(matches) > 1:
+                    self._debug(
+                        "Hardware speech auto found multiple devices for "
+                        f"{pattern}: {matches}; probing in order",
+                        debug.DebugLevel.WARNING,
+                        on_any_level=True,
+                    )
+                self._debug(
+                    f"Hardware speech auto candidate devices: {matches}",
+                    debug.DebugLevel.INFO,
+                    on_any_level=True,
+                )
+                devices.extend(matches)
+        return devices
 
     def _termios_baud_rate(self, baud_rate):
         baud_name = f"B{baud_rate}"
@@ -205,10 +375,23 @@ class hardware_serial_driver(speech_driver):
         value = max(0.0, min(1.0, value))
         return int(round(minimum + value * (maximum - minimum)))
 
-    def _debug(self, message, level):
+    def _format_bytes_preview(self, data, limit=32):
+        preview = data[:limit]
+        hex_preview = " ".join(f"{byte:02x}" for byte in preview)
+        ascii_preview = "".join(
+            chr(byte) if 0x20 <= byte <= 0x7E else "."
+            for byte in preview
+        )
+        suffix = "" if len(data) <= limit else " ..."
+        return (
+            f"hex=[{hex_preview}{suffix}] "
+            f"ascii=[{ascii_preview}{suffix}]"
+        )
+
+    def _debug(self, message, level, on_any_level=False):
         try:
             self.env["runtime"]["DebugManager"].write_debug_out(
-                message, level
+                message, level, on_any_level=on_any_level
             )
         except Exception:
             pass
diff --git a/src/fenrirscreenreader/speechDriver/litetalkDriver.py b/src/fenrirscreenreader/speechDriver/litetalkDriver.py
index 6dadeda2..75397781 100644
--- a/src/fenrirscreenreader/speechDriver/litetalkDriver.py
+++ b/src/fenrirscreenreader/speechDriver/litetalkDriver.py
@@ -11,6 +11,7 @@ from fenrirscreenreader.speechDriver.hardwareSerialDriver import (
 
 class driver(hardware_serial_driver):
     cancel_command = b"\x18"
+    hardware_probe_command = b"\x01I"
 
     def _speak_bytes(self, text):
         return self._clean_text(text).encode("ascii", errors="replace") + b"\r"
@@ -26,3 +27,10 @@ class driver(hardware_serial_driver):
 
     def _setting_command(self, value, command):
         return b"\x01" + str(value).encode("ascii") + command
+
+    def _is_hardware_probe_response(self, response):
+        response_text = response.decode("ascii", errors="ignore").lower()
+        return any(
+            token in response_text
+            for token in ("litetalk", "lite talk", "rpitalk", "doubletalk")
+        )
diff --git a/tests/unit/test_hardware_speech_drivers.py b/tests/unit/test_hardware_speech_drivers.py
index 8f5fce86..100b4947 100644
--- a/tests/unit/test_hardware_speech_drivers.py
+++ b/tests/unit/test_hardware_speech_drivers.py
@@ -1,6 +1,8 @@
 import os
 import select
+import termios
 import time
+from unittest.mock import ANY
 from unittest.mock import Mock
 
 import pytest
@@ -103,6 +105,280 @@ def test_litetalk_driver_writes_settings_and_cancel(serial_pair):
         speech_driver.shutdown()
 
 
+def test_auto_device_detection_includes_classic_serial(
+    monkeypatch, serial_pair
+):
+    master_fd, slave_name = serial_pair
+
+    def fake_glob(pattern):
+        if pattern == "/dev/ttyS*":
+            return [slave_name]
+        return []
+
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.glob.glob",
+        fake_glob,
+    )
+    speech_driver = litetalkDriver.driver()
+    speech_driver.initialize(build_environment("auto"))
+    try:
+        assert speech_driver.device == slave_name
+        speech_driver.speak("Serial")
+        assert read_available(master_fd, 9) == b"\x01ISerial\r"
+    finally:
+        speech_driver.shutdown()
+
+
+def test_configured_device_strips_inline_comment(serial_pair):
+    master_fd, slave_name = serial_pair
+    device_setting = f"{slave_name} # built-in serial port"
+    speech_driver = litetalkDriver.driver()
+    speech_driver.initialize(build_environment(device_setting))
+    try:
+        assert speech_driver.device == slave_name
+        speech_driver.speak("Specific")
+        assert read_available(master_fd, 9) == b"Specific\r"
+    finally:
+        speech_driver.shutdown()
+
+
+def test_auto_device_detection_prefers_probe_response(monkeypatch):
+    opened_ports = []
+    closed_ports = []
+    writes = {}
+
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.glob.glob",
+        lambda pattern: ["/dev/ttyUSB0", "/dev/ttyUSB1"]
+        if pattern == "/dev/ttyUSB*"
+        else [],
+    )
+
+    def fake_open(device, flags):
+        port = 100 + len(opened_ports)
+        opened_ports.append((device, port))
+        return port
+
+    def fake_write(port, data):
+        writes.setdefault(port, b"")
+        writes[port] += data
+        return len(data)
+
+    def fake_select(readable, writable, exceptional, timeout):
+        port = readable[0]
+        if port == 101 and writes.get(port) == b"\x01I":
+            return readable, writable, exceptional
+        return [], writable, exceptional
+
+    def fake_read(port, size):
+        if port == 101:
+            return b"RPItalk 1.3\r"
+        return b""
+
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.open",
+        fake_open,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.close",
+        lambda port: closed_ports.append(port),
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.write",
+        fake_write,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.read",
+        fake_read,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.select.select",
+        fake_select,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.termios.tcgetattr",
+        lambda port: [0, 0, 0, 0, 0, 0, [0] * 32],
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.termios.tcsetattr",
+        lambda port, when, attrs: None,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.tty.setraw",
+        lambda port: None,
+    )
+
+    speech_driver = litetalkDriver.driver()
+    speech_driver.initialize(build_environment("auto"))
+    try:
+        assert opened_ports == [("/dev/ttyUSB0", 100), ("/dev/ttyUSB1", 101)]
+        assert writes == {100: b"\x01I", 101: b"\x01I"}
+        assert closed_ports == [100]
+        assert speech_driver.device == "/dev/ttyUSB1"
+    finally:
+        speech_driver.shutdown()
+
+
+def test_auto_device_detection_falls_back_without_probe_response(
+    monkeypatch
+):
+    opened_ports = []
+    closed_ports = []
+
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.glob.glob",
+        lambda pattern: ["/dev/ttyUSB0", "/dev/ttyUSB1"]
+        if pattern == "/dev/ttyUSB*"
+        else [],
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.open",
+        lambda device, flags: opened_ports.append(device) or len(opened_ports),
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.close",
+        lambda port: closed_ports.append(port),
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.write",
+        lambda port, data: len(data),
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.select.select",
+        lambda readable, writable, exceptional, timeout: (
+            [],
+            writable,
+            exceptional,
+        ),
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.termios.tcgetattr",
+        lambda port: [0, 0, 0, 0, 0, 0, [0] * 32],
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.termios.tcsetattr",
+        lambda port, when, attrs: None,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.tty.setraw",
+        lambda port: None,
+    )
+
+    speech_driver = litetalkDriver.driver()
+    speech_driver.initialize(build_environment("auto"))
+    try:
+        assert opened_ports == ["/dev/ttyUSB0", "/dev/ttyUSB1"]
+        assert closed_ports == [2]
+        assert speech_driver.device == "/dev/ttyUSB0"
+    finally:
+        speech_driver.shutdown()
+
+
+def test_auto_device_detection_skips_termios_failures(monkeypatch):
+    opened_ports = []
+    closed_ports = []
+
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.glob.glob",
+        lambda pattern: ["/dev/ttyUSB0"]
+        if pattern == "/dev/ttyUSB*"
+        else ["/dev/ttyS0"]
+        if pattern == "/dev/ttyS*"
+        else [],
+    )
+
+    def fake_open(device, flags):
+        port = 100 + len(opened_ports)
+        opened_ports.append((device, port))
+        return port
+
+    def fake_tcgetattr(port):
+        if port == 101:
+            raise termios.error(5, "Input/output error")
+        return [0, 0, 0, 0, 0, 0, [0] * 32]
+
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.open",
+        fake_open,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.close",
+        lambda port: closed_ports.append(port),
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.write",
+        lambda port, data: len(data),
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.select.select",
+        lambda readable, writable, exceptional, timeout: (
+            [],
+            writable,
+            exceptional,
+        ),
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.termios.tcgetattr",
+        fake_tcgetattr,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.termios.tcsetattr",
+        lambda port, when, attrs: None,
+    )
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.tty.setraw",
+        lambda port: None,
+    )
+
+    speech_driver = litetalkDriver.driver()
+    speech_driver.initialize(build_environment("auto"))
+    try:
+        assert opened_ports == [("/dev/ttyUSB0", 100), ("/dev/ttyS0", 101)]
+        assert speech_driver.device == "/dev/ttyUSB0"
+    finally:
+        speech_driver.shutdown()
+
+
+def test_auto_device_detection_fails_when_no_serial_device(monkeypatch):
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.glob.glob",
+        lambda pattern: [],
+    )
+    speech_driver = litetalkDriver.driver()
+
+    with pytest.raises(RuntimeError, match="hardware speech device"):
+        speech_driver.initialize(build_environment("auto"))
+
+    debug_manager = speech_driver.env["runtime"]["DebugManager"]
+    debug_manager.write_debug_out.assert_called_with(
+        "Hardware speech device not found",
+        ANY,
+        on_any_level=True,
+    )
+
+
+def test_hardware_driver_retries_partial_serial_writes(monkeypatch):
+    speech_driver = litetalkDriver.driver()
+    speech_driver.env = build_environment("/dev/ttyUSB0")
+    speech_driver.serial_port = 12
+    written_chunks = []
+
+    def fake_write(port, data):
+        assert port == 12
+        chunk = data[:2]
+        written_chunks.append(chunk)
+        return len(chunk)
+
+    monkeypatch.setattr(
+        "fenrirscreenreader.speechDriver.hardwareSerialDriver.os.write",
+        fake_write,
+    )
+
+    speech_driver._write_bytes(b"abcdef", "speech")
+
+    assert written_chunks == [b"ab", b"cd", b"ef"]
+
+
 @pytest.mark.parametrize("driver_class", [doubletalkDriver, tripletalkDriver])
 def test_litetalk_compatible_alias_drivers(driver_class, serial_pair):
     speech_driver, master_fd = initialized_driver(driver_class, serial_pair)