diff --git a/docs/en/server/user-guide/configuration.md b/docs/en/server/user-guide/configuration.md
index 5726b47..f2aeb7b 100644
--- a/docs/en/server/user-guide/configuration.md
+++ b/docs/en/server/user-guide/configuration.md
@@ -33,7 +33,7 @@ With **systemd**, add to your unit file:
 ExecReload=/bin/kill -HUP $MAINPID
 ```
 
-**Reload applies:** `GLOBAL`, `REPORTS`, `ALIASES`, per-system settings, **new/removed SYSTEMS** (including `GENERATOR` expansion and new OpenBridge legs), and updated bind addresses (listener restart for that system only).
+**Reload applies:** `GLOBAL`, `REPORTS`, `ALIASES`, **`LOGGER.LOG_LEVEL`** (without process restart), per-system settings, **new/removed SYSTEMS** (including `GENERATOR` expansion and new OpenBridge legs), and updated bind addresses (listener restart for that system only).
 
 **Not reloaded:** `adn-voice.yaml` (separate 15 s loop), Python code, subscriber alias files (separate periodic reload). **BRIDGES** table is not rebuilt on reload — restart if bridge rules changed in a way that requires a full reset.
 
diff --git a/docs/en/server/user-guide/talker-alias.md b/docs/en/server/user-guide/talker-alias.md
index 8b8c8d0..45686bb 100644
--- a/docs/en/server/user-guide/talker-alias.md
+++ b/docs/en/server/user-guide/talker-alias.md
@@ -14,13 +14,18 @@ When enabled on a **MASTER** system, the server can:
 
 | Mode | Behaviour |
 |------|-----------|
-| **`both`** (default) | Pass through TA from the source hotspot/radio when all four `DMRA` blocks were received; otherwise inject from `subscriber_ids` + template. |
-| **`passthrough`** | Relay buffered `DMRA` only. |
-| **`inject`** | Always build TA from the configured template and alias data. |
+| **`both`** (default) | Prefer the source's own TA, fall back to inject. The server briefly waits (≈2 s after `VHEAD`) for the source's TA, decoded from either a valid MMDVM `DMRA` buffer or the embedded LC inside voice (FLCO 4–7). If found, it is **passed through unchanged**. If the source sent **no** TA within that window (e.g. a radio that does not support Talker Alias), ADN **injects** the configured template (embedded LC + `DMRA`). When the **source is an OpenBridge** system (which never carries TA), there is nothing to wait for, so ADN injects immediately at `VHEAD`. |
+| **`passthrough`** | Relay the source's TA only (buffered `DMRA` and the source embedded LC, both untouched). Never inject the template. |
+| **`inject`** | Always build TA from the configured template and alias data, overwriting the embedded LC and sending `DMRA` packets. |
 
 On **bridge forward** at voice header (`VHEAD`), the server sends four `DMRA` packets to each HBP target (**MASTER** peers or **PEER** upstream) once per stream, then forwards `DMRD` as usual.
 
-**MMDVMHost / DMRGateway (Pi-Star, WPSD):** stock MMDVMHost does **not** consume standalone downlink `DMRA` UDP; it decodes Talker Alias from **embedded LC inside `DMRD` voice** (FLCO 4–7). When TA is enabled, ADN injects TA into the embedded LC of voice bursts **B–E** (dtype 1–4) on **bridge forward** to HBP targets, in addition to optional standalone `DMRA` packets for clients that support them.
+**MMDVMHost / DMRGateway (Pi-Star, WPSD):** stock MMDVMHost does **not** consume standalone downlink `DMRA` UDP; it decodes Talker Alias from **embedded LC inside `DMRD` voice** (FLCO 4–7).
+
+The embedded LC of voice bursts **B–E** (dtype 1–4) is always rewritten with the **destination** group LC (re-encoded for the forwarded TGID, as legacy `bridge.py` does) — this is required for the receiving MMDVM to accept the voice; a stale embedded LC from the source TG causes packet loss. The Talker Alias is then **overlaid on alternate superframes**:
+
+- **`inject`** and the **`both`** no-source-TA fallback use the configured template.
+- **`passthrough`** and **`both`** with a source TA re-encode the source's own Talker Alias (decoded from its `DMRA`/embedded-voice blocks) and overlay that, so the radio's alias reaches the far side while the group LC stays correct for the new TG.
 
 On the **same MASTER**, when **`REPEAT`** copies group voice to other logged-in hotspots, the server also sends those four `DMRA` packets on `VHEAD` (excluding the transmitting peer). Bridge forwarding to the same system shares the same once-per-stream dedupe, so TA is not sent twice.
 
diff --git a/docs/es/server/user-guide/configuration.md b/docs/es/server/user-guide/configuration.md
index 7e6b8f2..7911d5c 100644
--- a/docs/es/server/user-guide/configuration.md
+++ b/docs/es/server/user-guide/configuration.md
@@ -33,7 +33,7 @@ Con **systemd**, en la unidad:
 ExecReload=/bin/kill -HUP $MAINPID
 ```
 
-**Se recarga:** `GLOBAL`, `REPORTS`, `ALIASES`, parámetros por system, **systems nuevos/eliminados** (incluida expansión `GENERATOR` y OBP nuevos), y cambios de IP/puerto (solo reinicia el listener de ese system).
+**Se recarga:** `GLOBAL`, `REPORTS`, `ALIASES`, **`LOGGER.LOG_LEVEL`** (sin reiniciar el proceso), parámetros por system, **systems nuevos/eliminados** (incluida expansión `GENERATOR` y OBP nuevos), y cambios de IP/puerto (solo reinicia el listener de ese system).
 
 **No se recarga:** `adn-voice.yaml` (loop aparte cada 15 s), código Python, ficheros de alias (recarga periódica). La tabla **BRIDGES** no se reconstruye — reinicia si cambiaste reglas de bridge que exijan reset completo.
 
diff --git a/pyproject.toml b/pyproject.toml
index 32b924e..c437104 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -27,5 +27,10 @@ docs = ["mkdocs>=1.6", "mkdocs-material>=9.5", "pymdown-extensions>=10.3"]
 [tool.setuptools.packages.find]
 where = ["src"]
 
+[tool.ruff.lint.per-file-ignores]
+# Entrypoints adjust sys.path before importing the package (intentional E402).
+"src/adn_server/main.py" = ["E402"]
+"src/adn_server/parrot_main.py" = ["E402"]
+
 [project.scripts]
 adn-server = "adn_server.main:main"
diff --git a/src/adn_server/application/bridge_use_cases.py b/src/adn_server/application/bridge_use_cases.py
index 8492991..2ec557d 100644
--- a/src/adn_server/application/bridge_use_cases.py
+++ b/src/adn_server/application/bridge_use_cases.py
@@ -41,7 +41,7 @@ from dmr_utils3.const import LC_OPT
 from ..domain import HBPF_DATA_SYNC, HBPF_SLT_VHEAD, HBPF_SLT_VTERM, STREAM_TO, bytes_3, bytes_4, int_id
 from ..domain.talker_alias import DMRA_BLOCK_COUNT
 from .ports import BridgeRouter
-from .talker_alias_use_cases import TalkerAliasUseCases
+from .talker_alias_use_cases import TalkerAliasUseCases, passthrough_complete, talker_alias_settings
 
 logger = logging.getLogger(__name__)
 
@@ -95,6 +95,7 @@ class BridgeUseCases:
         send_bcsq: Any = None,
         send_dmra_to_system: Any = None,
         get_dmra_blocks: Any = None,
+        call_later: Any = None,
     ) -> None:
         self._router = bridge_router
         self._config = config
@@ -105,7 +106,137 @@ class BridgeUseCases:
         self._send_bcsq = send_bcsq  # (system_name, tgid, stream_id) -> None; legacy OBP send_bcsq from router
         self._send_dmra_to_system = send_dmra_to_system
         self._get_dmra_blocks = get_dmra_blocks
+        self._call_later = call_later
         self._talker_alias = TalkerAliasUseCases(config)
+        # (source_system, stream_id) -> {rf_src, peer, targets, timer}
+        self._both_ta_wait: dict[tuple[str, bytes], dict[str, Any]] = {}
+
+    def _get_stream_dmra_blocks(self, source_system: str, stream_id: bytes) -> dict[int, bytes] | None:
+        if not self._get_dmra_blocks:
+            return None
+        return self._get_dmra_blocks(source_system, stream_id)
+
+    def _both_ta_key(self, source_system: str, stream_id: bytes) -> tuple[str, bytes]:
+        return (source_system, stream_id)
+
+    def _source_cannot_carry_ta(self, source_system: str) -> bool:
+        """OpenBridge carries no Talker Alias (no DMRA UDP nor embedded LC TA).
+
+        In `both` mode there is nothing to wait for, so inject the template right away
+        instead of deferring (OBP streams are often short and would expire first).
+        """
+        return self._config.get("SYSTEMS", {}).get(source_system, {}).get("MODE") == "OPENBRIDGE"
+
+    def _cancel_both_ta_wait(self, source_system: str, stream_id: bytes) -> None:
+        wait = self._both_ta_wait.pop(self._both_ta_key(source_system, stream_id), None)
+        if wait and wait.get("timer") and getattr(wait["timer"], "cancel", None):
+            try:
+                wait["timer"].cancel()
+            except Exception:
+                pass
+
+    def _register_both_ta_wait(
+        self,
+        source_system: str,
+        target_system: str,
+        rf_src: bytes,
+        stream_id: bytes,
+        source_peer: bytes,
+    ) -> None:
+        """Defer DMRA UDP + embed inject until MMDVM fragments arrive (both mode)."""
+        if not self._call_later:
+            return
+        key = self._both_ta_key(source_system, stream_id)
+        wait = self._both_ta_wait.setdefault(
+            key,
+            {"rf_src": rf_src, "peer": source_peer, "targets": set()},
+        )
+        wait["rf_src"] = rf_src
+        wait["peer"] = source_peer
+        wait["targets"].add(target_system)
+        if wait.get("timer"):
+            return
+        # Wait long enough to detect a slow source TA (MMDVM emits TA ~1 s in) before
+        # falling back to inject; passthrough is applied earlier via on_dmra_fragment_stored.
+        wait["timer"] = self._call_later(2.0, self._finalize_both_ta, source_system, stream_id)
+
+    def _finalize_both_ta(self, source_system: str, stream_id: bytes) -> None:
+        key = self._both_ta_key(source_system, stream_id)
+        wait = self._both_ta_wait.pop(key, None)
+        if not wait:
+            return
+        rf_src = wait["rf_src"]
+        peer = wait["peer"]
+        blocks = self._get_stream_dmra_blocks(source_system, stream_id)
+        # No source TA within the window: fall back to inject (template).
+        fallback_inject = not (blocks and passthrough_complete(blocks))
+        for target_system in wait["targets"]:
+            if not self._talker_alias.should_send_on_vhead(target_system, stream_id):
+                continue
+            self._send_talker_alias_to_target(
+                source_system, target_system, rf_src, stream_id, peer,
+                force=True, fallback_inject=fallback_inject,
+            )
+        self._apply_both_ta_embed(source_system, rf_src, stream_id, force_inject=fallback_inject)
+
+    def _apply_both_ta_embed(
+        self,
+        source_system: str,
+        rf_src: bytes,
+        stream_id: bytes,
+        *,
+        force_inject: bool = False,
+    ) -> None:
+        """Install embed LC state once the TA decision is known (passthrough or inject)."""
+        if not self._get_protocols:
+            return
+        for proto in self._get_protocols().values():
+            status = getattr(proto, "STATUS", None)
+            if not isinstance(status, dict):
+                continue
+            for st in status.values():
+                if not isinstance(st, dict):
+                    continue
+                if st.get("TX_STREAM_ID") != stream_id or st.get("TX_RFS") != rf_src:
+                    continue
+                if st.get("TX_TA_ON"):
+                    continue
+                target = st.get("_ta_target_system", source_system)
+                self._init_talker_alias_embed(
+                    st, source_system, target, rf_src, stream_id, force_inject=force_inject,
+                )
+
+    def on_dmra_fragment_stored(
+        self,
+        source_system: str,
+        peer_id: bytes,
+        rf_src: bytes,
+        stream_id: bytes,
+    ) -> None:
+        """Source TA may complete after VHEAD (DMRA UDP or decoded from voice).
+
+        Once the buffer is complete, overlay the source TA on the outgoing embedded LC
+        (passthrough/both). For `both` with a pending wait, also relay DMRA now and cancel
+        the inject fallback.
+        """
+        if talker_alias_settings(self._config, source_system)["mode"] not in ("both", "passthrough"):
+            return
+        blocks = self._get_stream_dmra_blocks(source_system, stream_id)
+        if not blocks or not passthrough_complete(blocks):
+            return
+        key = self._both_ta_key(source_system, stream_id)
+        wait = self._both_ta_wait.get(key)
+        if not wait:
+            self._apply_both_ta_embed(source_system, rf_src, stream_id)
+            return
+        wait["peer"] = peer_id
+        self._cancel_both_ta_wait(source_system, stream_id)
+        for target_system in wait["targets"]:
+            if self._talker_alias.should_send_on_vhead(target_system, stream_id):
+                self._send_talker_alias_to_target(
+                    source_system, target_system, rf_src, stream_id, peer_id, force=True,
+                )
+        self._apply_both_ta_embed(source_system, rf_src, stream_id)
 
     def _send_talker_alias_to_target(
         self,
@@ -114,6 +245,9 @@ class BridgeUseCases:
         rf_src: bytes,
         stream_id: bytes,
         source_peer: bytes,
+        *,
+        force: bool = False,
+        fallback_inject: bool = False,
     ) -> None:
         """Emit DMRA to an HBP target on VHEAD (once per target stream)."""
         if not self._send_dmra_to_system:
@@ -123,12 +257,28 @@ class BridgeUseCases:
             return
         if not self._talker_alias.should_send_on_vhead(target_system, stream_id):
             return
+        if (
+            not force
+            and talker_alias_settings(self._config, source_system)["mode"] == "both"
+            and not (self._get_stream_dmra_blocks(source_system, stream_id) and passthrough_complete(
+                self._get_stream_dmra_blocks(source_system, stream_id) or {}
+            ))
+        ):
+            if self._source_cannot_carry_ta(source_system):
+                # OBP source can never supply TA: inject the template immediately.
+                fallback_inject = True
+            else:
+                self._register_both_ta_wait(
+                    source_system, target_system, rf_src, stream_id, source_peer,
+                )
+                return
         packets = self._talker_alias.packets_for_stream(
             source_system,
             rf_src,
             stream_id,
             self._get_dmra_blocks,
             target_system=target_system,
+            fallback_inject=fallback_inject,
         )
         if not packets:
             return
@@ -138,6 +288,7 @@ class BridgeUseCases:
         except Exception as e:
             logger.warning("(ROUTER) send_dmra_to_system %s failed: %s", target_system, e)
             return
+        self._talker_alias.mark_dmra_sent(target_system, stream_id)
         sid = int_id(stream_id)
         if peer_count:
             logger.debug(
@@ -164,10 +315,13 @@ class BridgeUseCases:
 
     def clear_talker_alias_stream(self, system_name: str, stream_id: bytes) -> None:
         """Release per-stream TA dedupe state after VTERM."""
+        self._cancel_both_ta_wait(system_name, stream_id)
         self._talker_alias.clear_stream(system_name, stream_id)
         if not self._get_protocols:
             return
         proto = self._get_protocols().get(system_name)
+        if proto is not None and hasattr(proto, "clear_ta_stream_buffer"):
+            proto.clear_ta_stream_buffer(stream_id)
         if proto is None:
             return
         status = getattr(proto, "STATUS", None)
@@ -185,8 +339,31 @@ class BridgeUseCases:
         target_system: str,
         rf_src: bytes,
         stream_id: bytes,
+        *,
+        force_inject: bool = False,
     ) -> None:
-        """Prepare per-stream embedded TA state for DMRD voice injection."""
+        """Prepare per-stream embedded TA state for DMRD voice injection.
+
+        The group LC is always rewritten for the destination TG by ``_rewrite_embed_lc``;
+        here we only set ``TX_TA_EMB`` when a Talker Alias should be overlaid. The source
+        TA (passthrough/both) is re-encoded from its decoded DMRA/voice blocks once they
+        arrive; the template is used for ``inject`` and the ``both`` fallback. If no TA is
+        available yet (e.g. at VHEAD, before the source TA has been decoded), TX_TA_EMB is
+        left unset and only the destination group LC is emitted until it becomes available.
+        """
+        st["_ta_source_system"] = source_system
+        st["_ta_target_system"] = target_system
+        settings = talker_alias_settings(self._config, source_system)
+        if not settings["enabled"]:
+            return
+        target_mode = self._config.get("SYSTEMS", {}).get(target_system, {}).get("MODE")
+        if (
+            settings["mode"] == "both"
+            and self._source_cannot_carry_ta(source_system)
+            and target_mode in ("MASTER", "PEER")
+        ):
+            # OBP source can never supply TA: inject the template immediately.
+            force_inject = True
         st.pop("TX_TA_EMB", None)
         st.pop("TX_TA_PHASE", None)
         st.pop("TX_TA_ON", None)
@@ -196,6 +373,7 @@ class BridgeUseCases:
             stream_id,
             self._get_dmra_blocks,
             target_system=target_system,
+            fallback_inject=force_inject,
         )
         if emblcs:
             st["TX_TA_EMB"], st["TX_TA_BLOCK_COUNT"] = emblcs
@@ -203,8 +381,21 @@ class BridgeUseCases:
             # First B1–B4 cycle carries group LC; TA on the next cycle.
             st["TX_TA_ON"] = False
 
-    def _rewrite_embed_lc(self, dmrbits: bitarray, st: dict[str, Any], dtype_vseq: int, emb_key: str) -> None:
-        """Replace embedded LC on voice bursts B–E; alternate group LC and TA blocks."""
+    def _rewrite_embed_lc(
+        self,
+        dmrbits: bitarray,
+        st: dict[str, Any],
+        dtype_vseq: int,
+        emb_key: str,
+    ) -> None:
+        """Replace embedded LC on voice bursts B–E (legacy bridge.py parity).
+
+        Group-call superframes always carry the **destination** group LC (``emb_key``),
+        re-encoded for the rewritten TGID — this is required for the voice to be accepted
+        by the receiving MMDVM (a mismatched embedded LC causes packet loss). When a
+        Talker Alias is available (``TX_TA_EMB``: injected template or the source TA
+        re-encoded from its DMRA/voice blocks) it is overlaid on alternate superframes.
+        """
         if dtype_vseq not in (1, 2, 3, 4):
             return
         ta_emb = st.get("TX_TA_EMB")
diff --git a/src/adn_server/application/playback_use_cases.py b/src/adn_server/application/playback_use_cases.py
index af07732..e6761b0 100644
--- a/src/adn_server/application/playback_use_cases.py
+++ b/src/adn_server/application/playback_use_cases.py
@@ -65,6 +65,7 @@ class PlaybackUseCases:
         self._playback_packets: list[bytes] = []
         self._playback_index = 0
         self._playback_stream_id = b""
+        self._playback_ta_from_stream = b""
         self._delay_call: DelayedCall | None = None
         self._packet_call: DelayedCall | None = None
         self._idle_call: DelayedCall | None = None
@@ -197,6 +198,14 @@ class PlaybackUseCases:
         dst_id: bytes,
     ) -> None:
         self.CALL_DATA.append(data)
+        if proto is not None and hasattr(proto, "store_ta_from_voice_burst") and len(data) >= 53:
+            bits = data[15]
+            frame_type = (bits & 0x30) >> 4
+            vseq = bits & 0xF
+            if frame_type != HBPF_DATA_SYNC and vseq in (1, 2, 3, 4):
+                proto.store_ta_from_voice_burst(
+                    peer_id, rf_src, self._record_stream, vseq, data[20:53],
+                )
         self._last_record_time = pkt_time
         self._record_ctx = {
             "slot": slot,
@@ -263,6 +272,7 @@ class PlaybackUseCases:
         ctx = self._record_ctx
         slot = ctx.get("slot", 1)
         recorded = self._ensure_vterm(list(self.CALL_DATA), slot)
+        self._playback_ta_from_stream = self._record_stream
         self._reset_recording_state(slot)
         if not recorded:
             return
@@ -363,6 +373,9 @@ class PlaybackUseCases:
             return
 
         self._playback_stream_id = bytes_4(randint(0x00, 0xFFFFFFFF))
+        if proto is not None and self._playback_ta_from_stream and hasattr(proto, "copy_ta_stream_buffer"):
+            proto.copy_ta_stream_buffer(self._playback_ta_from_stream, self._playback_stream_id)
+        self._playback_ta_from_stream = b""
         logger.info(
             "(%s) *START  PLAYBACK* STREAM ID: %s SUB: %s REPEATER: %s TGID %s, TS %s, Duration: %.2f",
             self._system,
@@ -424,5 +437,6 @@ class PlaybackUseCases:
         self._playback_packets = []
         self._playback_index = 0
         self._playback_stream_id = b""
+        self._playback_ta_from_stream = b""
         self._delay_call = None
         self._packet_call = None
diff --git a/src/adn_server/application/talker_alias_use_cases.py b/src/adn_server/application/talker_alias_use_cases.py
index 0a8c925..cd8769c 100644
--- a/src/adn_server/application/talker_alias_use_cases.py
+++ b/src/adn_server/application/talker_alias_use_cases.py
@@ -13,12 +13,16 @@ from ..domain.talker_alias import (
     DMRA_BLOCK_COUNT,
     DMRA_PAYLOAD_LEN,
     buffer_from_blocks,
+    buffer_from_wire_blocks,
     build_dmra_packet,
     build_dmra_packets,
-    decode_7bit,
+    decode_ta_from_blocks,
     encode_talker_alias_emblc,
     encode_talker_alias_emblc_from_blocks,
+    required_ta_block_count,
+    talker_alias_decode_complete,
     truncate_talker_alias,
+    is_ta_header_byte,
 )
 
 logger = logging.getLogger(__name__)
@@ -77,18 +81,40 @@ def format_talker_alias_text(config: dict[str, Any], rf_src: bytes) -> str:
     return truncate_talker_alias(text)
 
 
+def _passthrough_buf_ready(blocks: dict[int, bytes], buf: bytes) -> bool:
+    need = required_ta_block_count(buf)
+    if need < 1 or need > DMRA_BLOCK_COUNT:
+        return False
+    if not all(i in blocks and len(blocks[i]) >= DMRA_PAYLOAD_LEN for i in range(need)):
+        return False
+    return talker_alias_decode_complete(buf)
+
+
 def passthrough_complete(blocks: dict[int, bytes]) -> bool:
-    """True when all four TA blocks were received."""
-    return all(i in blocks and len(blocks[i]) >= DMRA_PAYLOAD_LEN for i in range(DMRA_BLOCK_COUNT))
+    """True when required TA blocks (1–4) are present and fully decode."""
+    if not blocks:
+        return False
+    buf = buffer_from_wire_blocks(blocks)
+    if _passthrough_buf_ready(blocks, buf):
+        return True
+    if blocks.get(0) and is_ta_header_byte(blocks[0][0]):
+        return _passthrough_buf_ready(blocks, buffer_from_blocks(blocks))
+    return False
 
 
 def passthrough_packets_from_blocks(rf_src: bytes, blocks: dict[int, bytes]) -> list[bytes]:
-    """Rebuild four DMRA packets from buffered block payloads."""
-    packets: list[bytes] = []
-    for block_id in range(DMRA_BLOCK_COUNT):
-        payload = blocks.get(block_id, b"\x00" * DMRA_PAYLOAD_LEN)
-        packets.append(build_dmra_packet(rf_src, block_id, payload))
-    return packets
+    """Rebuild DMRA packets from buffered wire payloads (only blocks needed for TA)."""
+    if passthrough_complete(blocks):
+        buf = buffer_from_wire_blocks(blocks)
+        if not talker_alias_decode_complete(buf) and blocks.get(0) and is_ta_header_byte(blocks[0][0]):
+            buf = buffer_from_blocks(blocks)
+        count = required_ta_block_count(buf)
+        return [
+            build_dmra_packet(rf_src, block_id, blocks[block_id])
+            for block_id in range(count)
+            if block_id in blocks
+        ]
+    return []
 
 
 class TalkerAliasUseCases:
@@ -104,11 +130,10 @@ class TalkerAliasUseCases:
         self._embed_logged.discard((system_name, stream_id))
 
     def should_send_on_vhead(self, target_system: str, stream_id: bytes) -> bool:
-        key = (target_system, stream_id)
-        if key in self._sent_streams:
-            return False
-        self._sent_streams.add(key)
-        return True
+        return (target_system, stream_id) not in self._sent_streams
+
+    def mark_dmra_sent(self, target_system: str, stream_id: bytes) -> None:
+        self._sent_streams.add((target_system, stream_id))
 
     def packets_for_stream(
         self,
@@ -118,6 +143,7 @@ class TalkerAliasUseCases:
         get_passthrough_blocks: Any,
         *,
         target_system: str | None = None,
+        fallback_inject: bool = False,
     ) -> list[bytes] | None:
         """Return DMRA packets to send on VHEAD, or None if TA disabled / nothing to send."""
         settings = talker_alias_settings(self._config, source_system)
@@ -131,7 +157,7 @@ class TalkerAliasUseCases:
         if mode == "passthrough":
             if not have_passthrough:
                 return None
-            text = decode_7bit(buffer_from_blocks(blocks))
+            text = decode_ta_from_blocks(blocks)
             logger.debug(
                 "(%s) *TALKER ALIAS* passthrough '%s' via %s -> %s stream %s",
                 source_system, text, via, target, int_id(stream_id),
@@ -144,20 +170,28 @@ class TalkerAliasUseCases:
                 source_system, text, via, target, int_id(stream_id),
             )
             return build_dmra_packets(rf_src, text)
-        # both
+        # both: prefer the source's own TA. If a valid MMDVM DMRA buffer arrived,
+        # relay it. Otherwise the source's embedded LC (e.g. MMDVM voice) is passed
+        # through unchanged in the DMRD voice, so do NOT inject a template here.
         if have_passthrough:
-            text = decode_7bit(buffer_from_blocks(blocks))
+            text = decode_ta_from_blocks(blocks)
             logger.debug(
                 "(%s) *TALKER ALIAS* passthrough '%s' via %s -> %s stream %s",
                 source_system, text, via, target, int_id(stream_id),
             )
             return passthrough_packets_from_blocks(rf_src, blocks)
-        text = format_talker_alias_text(self._config, rf_src)
+        if fallback_inject:
+            text = format_talker_alias_text(self._config, rf_src)
+            logger.debug(
+                "(%s) *TALKER ALIAS* inject '%s' (no source TA) via %s -> %s stream %s",
+                source_system, text, via, target, int_id(stream_id),
+            )
+            return build_dmra_packets(rf_src, text)
         logger.debug(
-            "(%s) *TALKER ALIAS* inject '%s' via %s -> %s stream %s",
-            source_system, text, via, target, int_id(stream_id),
+            "(%s) *TALKER ALIAS* passthrough (source embedded TA) via %s -> %s stream %s",
+            source_system, via, target, int_id(stream_id),
         )
-        return build_dmra_packets(rf_src, text)
+        return None
 
     def embedded_emblc_for_stream(
         self,
@@ -167,43 +201,41 @@ class TalkerAliasUseCases:
         get_passthrough_blocks: Any,
         *,
         target_system: str | None = None,
+        fallback_inject: bool = False,
     ) -> tuple[list[dict[int, Any]], int] | None:
-        """Return (encode_emblc dicts, block count 1–4) for embedded TA in DMRD, or None."""
+        """Return (encode_emblc dicts, block count 1–4) for the embedded TA to overlay, or None.
+
+        The group LC is rewritten separately for the destination TG; this only supplies the
+        Talker Alias overlaid on alternate superframes:
+
+        - **passthrough / both with source TA:** re-encode the source's decoded TA blocks
+          (round-trips losslessly via the fixed ``encode_emblc``).
+        - **inject / both fallback:** the configured template.
+        - otherwise (no TA available yet): ``None`` — only the destination group LC is sent.
+        """
         settings = talker_alias_settings(self._config, source_system)
         if not settings["enabled"]:
             return None
+        mode = settings["mode"]
         target = target_system or source_system
         via = "repeat" if source_system == target else "bridge"
-        mode = settings["mode"]
         log_key = (source_system, stream_id)
-        log_embed = log_key not in self._embed_logged
 
-        def _log_embed(text: str, passthrough: bool) -> None:
-            if not log_embed:
+        def _log(text: str, suffix: str) -> None:
+            if log_key in self._embed_logged:
                 return
             self._embed_logged.add(log_key)
-            kind = "passthrough" if passthrough else "inject"
             logger.debug(
-                "(%s) *TALKER ALIAS* embed %s '%s' via %s -> %s stream %s",
-                source_system, kind, text, via, target, int_id(stream_id),
+                "(%s) *TALKER ALIAS* embed inject '%s'%s via %s -> %s stream %s",
+                source_system, text, suffix, via, target, int_id(stream_id),
             )
 
         blocks = get_passthrough_blocks(source_system, stream_id) if get_passthrough_blocks else None
-        have_passthrough = bool(blocks and passthrough_complete(blocks))
-        if mode == "passthrough":
-            if not have_passthrough:
-                return None
-            text = decode_7bit(buffer_from_blocks(blocks))
-            _log_embed(text, True)
-            return encode_talker_alias_emblc_from_blocks(blocks)
-        if mode == "inject":
-            text = format_talker_alias_text(self._config, rf_src)
-            _log_embed(text, False)
-            return encode_talker_alias_emblc(text)
-        if have_passthrough:
-            text = decode_7bit(buffer_from_blocks(blocks))
-            _log_embed(text, True)
+        if mode in ("passthrough", "both") and blocks and passthrough_complete(blocks):
+            _log(decode_ta_from_blocks(blocks), " (source TA)")
             return encode_talker_alias_emblc_from_blocks(blocks)
-        text = format_talker_alias_text(self._config, rf_src)
-        _log_embed(text, False)
-        return encode_talker_alias_emblc(text)
+        if mode == "inject" or fallback_inject:
+            suffix = "" if mode == "inject" else " (no source TA)"
+            _log(format_talker_alias_text(self._config, rf_src), suffix)
+            return encode_talker_alias_emblc(format_talker_alias_text(self._config, rf_src))
+        return None
diff --git a/src/adn_server/domain/talker_alias.py b/src/adn_server/domain/talker_alias.py
index 5044840..0d12fe0 100644
--- a/src/adn_server/domain/talker_alias.py
+++ b/src/adn_server/domain/talker_alias.py
@@ -121,7 +121,7 @@ def required_ta_block_count(buf: bytes) -> int:
 
 
 def buffer_from_blocks(blocks: dict[int, bytes]) -> bytes:
-    """Merge up to four block payloads into a 28-byte buffer."""
+    """Merge up to four block payloads into a 28-byte buffer (ADN inject layout)."""
     buf = bytearray(DMRA_BUF_LEN)
     for block_id, payload in blocks.items():
         if 0 <= block_id < DMRA_BLOCK_COUNT and payload:
@@ -130,6 +130,46 @@ def buffer_from_blocks(blocks: dict[int, bytes]) -> bytes:
     return bytes(buf)
 
 
+def buffer_from_wire_blocks(blocks: dict[int, bytes]) -> bytes:
+    """Merge MMDVMHost DMRA wire payloads into the 28-byte TA buffer.
+
+    ``writeTalkerAlias`` copies ``data + 2`` (7 bytes) from the 9-byte embedded LC;
+    ``CDMRTA::add(blockId, data + 2, 7)`` stores those at ``m_buf[blockId * 7]``.
+    The UDP payload is therefore the same layout as ``buffer_from_blocks``.
+    """
+    return buffer_from_blocks(blocks)
+
+
+def is_ta_header_byte(byte0: int) -> bool:
+    """True if byte looks like ETSI TA header (UTF-8/ISO formats; not 7-bit)."""
+    if (byte0 & 1) != 0:
+        return False
+    fmt = (byte0 >> 6) & 0x03
+    size = (byte0 >> 1) & 0x1F
+    return fmt in (1, 2) and 1 <= size <= TALKER_ALIAS_MAX_LEN
+
+
+def talker_alias_decode_complete(buf: bytes) -> bool:
+    """True when TA buffer decodes with expected size (MMDVM DMRTA parity)."""
+    if len(buf) < 1 or not is_ta_header_byte(buf[0]):
+        return False
+    ta_size = (buf[0] >> 1) & 0x1F
+    text = decode_ta(buf).rstrip("\x00").strip()
+    return len(text) >= ta_size
+
+
+def decode_ta_from_blocks(blocks: dict[int, bytes]) -> str:
+    """Decode TA from buffered DMRA blocks (MMDVM wire layout, then ADN layout)."""
+    buf = buffer_from_wire_blocks(blocks)
+    if talker_alias_decode_complete(buf):
+        return decode_ta(buf).rstrip("\x00").strip()
+    if blocks.get(0) and is_ta_header_byte(blocks[0][0]):
+        buf = buffer_from_blocks(blocks)
+        if talker_alias_decode_complete(buf):
+            return decode_ta(buf).rstrip("\x00").strip()
+    return ""
+
+
 def build_dmra_packets(rf_src: bytes, text: str) -> list[bytes]:
     """Build HBP DMRA packets (1–4) for server injection."""
     rf = rf_src[:3] if len(rf_src) >= 3 else rf_src.ljust(3, b"\x00")[:3]
@@ -151,13 +191,81 @@ def build_dmra_packet(rf_src: bytes, block_id: int, payload: bytes) -> bytes:
 
 
 def parse_dmra_packet(data: bytes) -> tuple[bytes, int, bytes] | None:
-    """Parse DMRA packet; returns (rf_src, block_id, payload) or None."""
+    """Parse MMDVMHost HBP DMRA (15 bytes), per ``DMRNetwork::writeTalkerAlias``.
+
+    Layout: ``DMRA`` | src_id[24-bit BE @4–6] | type @7 (0–3) | memcpy(@8, data+2, 7).
+
+    ``data`` is the 9-byte embedded LC (``CDMREmbeddedData::getRawData``); wire bytes equal
+    ``CDMRTA::add(blockId, data+2, 7)`` → ``m_buf[blockId*7 : blockId*7+7]``.
+    """
     if len(data) < DMRA_PACKET_LEN or data[:4] != DMRA_OPCODE:
         return None
-    rf_src = data[4:7]
     block_id = data[7]
-    payload = data[8:15]
-    return rf_src, block_id, payload
+    if block_id > 3:
+        return None
+    return data[4:7], block_id, data[8:15]
+
+
+def store_ta_block(blocks: dict[int, bytes], block_id: int, payload: bytes) -> bool:
+    """Store one wire fragment (7 bytes at ``m_buf[block_id*7]``)."""
+    if block_id < 0 or block_id > 3:
+        return False
+    blocks[block_id] = payload[:DMRA_PAYLOAD_LEN]
+    return True
+
+
+def store_ta_from_embed_lc(blocks: dict[int, bytes], block_id: int, lc9: bytes) -> bool:
+    """Store from 9-byte embedded LC (MMDVM ``data`` in ``DMRSlot`` TA FLCO 4–7 path)."""
+    if block_id < 0 or block_id > 3 or len(lc9) < 9:
+        return False
+    return store_ta_block(blocks, block_id, lc9[2:9])
+
+
+def talker_alias_block_id_from_lc(lc: bytes) -> int | None:
+    """FLCO 4–7 (MMDVM TALKER_ALIAS_HEADER..BLOCK3) → block index 0–3, else None."""
+    if len(lc) < 9:
+        return None
+    flco = lc[0]
+    if flco < FLCO_TALKER_ALIAS_HEADER or flco > FLCO_TALKER_ALIAS_BLOCK3:
+        return None
+    return flco - FLCO_TALKER_ALIAS_HEADER
+
+
+def try_buffer_ta_from_voice_fragments(
+    acc: dict[int, "bitarray"],
+    vseq: int,
+    dmrpkt: bytes,
+    blocks: dict[int, bytes],
+) -> bool:
+    """Reassemble embedded LC across voice bursts B–E (vseq 1–4) and store if it is a TA.
+
+    ``dmr_utils3.bptc.decode_emblc`` is correct on properly FEC-encoded input (the encode
+    helper is the buggy one), so a TA sent by a real radio decodes losslessly here.
+    Returns True when a TA block was decoded and stored.
+    """
+    from dmr_utils3 import bptc, decode
+
+    if vseq not in (1, 2, 3, 4) or len(dmrpkt) < 33:
+        return False
+    try:
+        embed = decode.voice(dmrpkt)["EMBED"]
+    except Exception:
+        return False
+    if vseq == 1:
+        acc.clear()
+    acc[vseq] = embed
+    if not all(i in acc for i in (1, 2, 3, 4)):
+        return False
+    try:
+        lc = bptc.decode_emblc(acc[1] + acc[2] + acc[3] + acc[4])
+    except Exception:
+        acc.clear()
+        return False
+    acc.clear()
+    block_id = talker_alias_block_id_from_lc(lc)
+    if block_id is None:
+        return False
+    return store_ta_from_embed_lc(blocks, block_id, lc)
 
 
 def talker_alias_lc_bytes(block_id: int, payload7: bytes) -> bytes:
@@ -167,23 +275,90 @@ def talker_alias_lc_bytes(block_id: int, payload7: bytes) -> bytes:
     return bytes([FLCO_TALKER_ALIAS_HEADER + block, 0x00]) + payload
 
 
+def encode_emblc(_lc: bytes) -> dict[int, bitarray]:
+    """Embedded-LC BPTC encode with the ``dmr_utils3`` bug fixed.
+
+    Upstream ``dmr_utils3.bptc.encode_emblc`` builds segment D's second row from
+    ``_binlc[24]`` instead of ``_binlc[25]`` (duplicating bit 24, dropping bit 25), which
+    corrupts one bit of the embedded LC (e.g. injected ``Rodrigo`` decodes as ``Rodrigg``).
+    This is a faithful copy with that single index corrected, so injected Talker Alias
+    round-trips losslessly through ``decode_emblc`` and on real radios.
+    """
+    from bitarray import bitarray
+    from dmr_utils3 import crc, hamming
+
+    _csum = crc.csum5(_lc)
+    _binlc = bitarray(endian="big")
+    _binlc.frombytes(_lc)
+    _binlc.insert(32, _csum[0])
+    _binlc.insert(43, _csum[1])
+    _binlc.insert(54, _csum[2])
+    _binlc.insert(65, _csum[3])
+    _binlc.insert(76, _csum[4])
+    for index in range(0, 112, 16):
+        for hindex, hbit in zip(
+            range(index + 11, index + 16), hamming.enc_16114(_binlc[index:index + 11])
+        ):
+            _binlc.insert(hindex, hbit)
+    for index in range(0, 16):
+        _binlc.insert(
+            index + 112,
+            _binlc[index] ^ _binlc[index + 16] ^ _binlc[index + 32] ^ _binlc[index + 48]
+            ^ _binlc[index + 64] ^ _binlc[index + 80] ^ _binlc[index + 96],
+        )
+
+    def _seg(rows: tuple[tuple[int, ...], ...]) -> bitarray:
+        out = bitarray(endian="big")
+        for row in rows:
+            out.extend([_binlc[i] for i in row])
+        return out
+
+    emblc_b = _seg((
+        (0, 16, 32, 48, 64, 80, 96, 112),
+        (1, 17, 33, 49, 65, 81, 97, 113),
+        (2, 18, 34, 50, 66, 82, 98, 114),
+        (3, 19, 35, 51, 67, 83, 99, 115),
+    ))
+    emblc_c = _seg((
+        (4, 20, 36, 52, 68, 84, 100, 116),
+        (5, 21, 37, 53, 69, 85, 101, 117),
+        (6, 22, 38, 54, 70, 86, 102, 118),
+        (7, 23, 39, 55, 71, 87, 103, 119),
+    ))
+    emblc_d = _seg((
+        (8, 24, 40, 56, 72, 88, 104, 120),
+        (9, 25, 41, 57, 73, 89, 105, 121),  # fixed: bit 25 (dmr_utils3 wrongly uses 24)
+        (10, 26, 42, 58, 74, 90, 106, 122),
+        (11, 27, 43, 59, 75, 91, 107, 123),
+    ))
+    emblc_e = _seg((
+        (12, 28, 44, 60, 76, 92, 108, 124),
+        (13, 29, 45, 61, 77, 93, 109, 125),
+        (14, 30, 46, 62, 78, 94, 110, 126),
+        (15, 31, 47, 63, 79, 95, 111, 127),
+    ))
+    return {1: emblc_b, 2: emblc_c, 3: emblc_d, 4: emblc_e}
+
+
 def encode_talker_alias_emblc(text: str) -> tuple[list[dict[int, bitarray]], int]:
     """Embedded-LC dicts for TA blocks 0..N-1 and block count N (1–4)."""
-    from dmr_utils3 import bptc
-
     encoded = encode_utf8(text)
     blocks = blocks_from_buffer(encoded)
     count = required_ta_block_count(encoded)
-    emblcs = [bptc.encode_emblc(talker_alias_lc_bytes(i, blocks[i])) for i in range(count)]
+    emblcs = [encode_emblc(talker_alias_lc_bytes(i, blocks[i])) for i in range(count)]
     return emblcs, count
 
 
 def encode_talker_alias_emblc_from_blocks(blocks: dict[int, bytes]) -> tuple[list[dict[int, bitarray]], int]:
-    """Build embedded TA LC dicts from buffered DMRA payloads."""
-    from dmr_utils3 import bptc
-
-    buf = buffer_from_blocks(blocks)
-    buf_blocks = blocks_from_buffer(buf)
+    """Build embedded TA LC dicts from buffered DMRA wire payloads."""
+    buf = buffer_from_wire_blocks(blocks)
     count = required_ta_block_count(buf)
-    emblcs = [bptc.encode_emblc(talker_alias_lc_bytes(i, buf_blocks[i])) for i in range(count)]
+    if not talker_alias_decode_complete(buf) and blocks.get(0) and is_ta_header_byte(blocks[0][0]):
+        buf = buffer_from_blocks(blocks)
+        count = required_ta_block_count(buf)
+    emblcs = [
+        encode_emblc(talker_alias_lc_bytes(i, blocks[i]))
+        for i in range(count)
+        if i in blocks
+    ]
     return emblcs, count
diff --git a/src/adn_server/infrastructure/__init__.py b/src/adn_server/infrastructure/__init__.py
index 1d67bd1..439606a 100644
--- a/src/adn_server/infrastructure/__init__.py
+++ b/src/adn_server/infrastructure/__init__.py
@@ -22,6 +22,6 @@
 ###############################################################################
 
 from .config_loader import YamlConfigLoader
-from .logging_config import reopen_file_handlers, setup_logging
+from .logging_config import reapply_log_level, reopen_file_handlers, setup_logging
 
-__all__ = ["YamlConfigLoader", "reopen_file_handlers", "setup_logging"]
+__all__ = ["YamlConfigLoader", "reapply_log_level", "reopen_file_handlers", "setup_logging"]
diff --git a/src/adn_server/infrastructure/config_reload.py b/src/adn_server/infrastructure/config_reload.py
index ff5b055..69acc51 100644
--- a/src/adn_server/infrastructure/config_reload.py
+++ b/src/adn_server/infrastructure/config_reload.py
@@ -12,6 +12,7 @@ from typing import Any, Callable
 
 from ..domain.errors import ConfigError
 from .config_loader import YamlConfigLoader
+from .logging_config import reapply_log_level
 from .config_normalizer import (
     apply_talker_alias_defaults,
     ensure_system_runtime_config,
@@ -146,6 +147,9 @@ def reload_server_config(
         raise
 
     merge_top_level_config(config, incoming)
+    if "LOGGER" in incoming:
+        level_name = reapply_log_level(config.get("LOGGER", {}))
+        log.info("(CONFIG-RELOAD) LOG_LEVEL applied: %s", level_name)
 
     old_systems = dict(config.get("SYSTEMS", {}))
     new_systems = incoming.get("SYSTEMS", {})
diff --git a/src/adn_server/infrastructure/logging_config.py b/src/adn_server/infrastructure/logging_config.py
index ec79e45..9a26557 100644
--- a/src/adn_server/infrastructure/logging_config.py
+++ b/src/adn_server/infrastructure/logging_config.py
@@ -72,6 +72,27 @@ def reopen_file_handlers(logger: logging.Logger | None = None) -> int:
     return count
 
 
+def reapply_log_level(log_config: dict[str, Any]) -> str:
+    """Apply LOGGER.LOG_LEVEL after SIGHUP reload (handlers unchanged).
+
+    Returns the level name applied (e.g. ``DEBUG``).
+    """
+    if not logging_enabled(log_config):
+        level = logging.CRITICAL
+    else:
+        level = getattr(logging, (log_config.get("LOG_LEVEL", "INFO")).upper(), logging.INFO)
+    log_name = log_config.get("LOG_NAME", "ADN")
+    root = logging.getLogger()
+    root.setLevel(level)
+    app_logger = logging.getLogger(log_name)
+    app_logger.setLevel(level)
+    for handler in root.handlers:
+        handler.setLevel(level)
+    for handler in app_logger.handlers:
+        handler.setLevel(level)
+    return logging.getLevelName(level)
+
+
 def setup_logging(log_config: dict[str, Any]) -> logging.Logger:
     """Configure logging from CONFIG['LOGGER']. Returns application logger."""
     log_name = log_config.get("LOG_NAME", "ADN")
diff --git a/src/adn_server/infrastructure/twisted_adapters/udp_hbp.py b/src/adn_server/infrastructure/twisted_adapters/udp_hbp.py
index 49100be..a7c27f3 100644
--- a/src/adn_server/infrastructure/twisted_adapters/udp_hbp.py
+++ b/src/adn_server/infrastructure/twisted_adapters/udp_hbp.py
@@ -43,7 +43,13 @@ from twisted.internet import reactor, task
 from twisted.internet.protocol import DatagramProtocol
 
 from ...domain import bytes_4, int_id
-from ...domain.talker_alias import DMRA_PACKET_LEN, parse_dmra_packet
+from ...domain.talker_alias import (
+    DMRA_PACKET_LEN,
+    decode_ta_from_blocks,
+    parse_dmra_packet,
+    store_ta_block,
+    try_buffer_ta_from_voice_fragments,
+)
 from ..hbp_constants import (
     BC,
     BCKA,
@@ -146,6 +152,7 @@ class HBPProtocol(DatagramProtocol):
         on_obp_bcsq_received: Callable[[str, bytes, bytes], None] | None = None,
         on_talker_alias_local_repeat: Callable[[str, bytes, bytes, bytes], None] | None = None,
         on_talker_alias_stream_end: Callable[[str, bytes], None] | None = None,
+        on_dmra_fragment_stored: Callable[[str, bytes, bytes, bytes], None] | None = None,
     ) -> None:
         self._CONFIG = config
         self._system = system_name
@@ -161,6 +168,7 @@ class HBPProtocol(DatagramProtocol):
         self._on_obp_bcsq_received = on_obp_bcsq_received
         self._on_talker_alias_local_repeat = on_talker_alias_local_repeat
         self._on_talker_alias_stream_end = on_talker_alias_stream_end
+        self._on_dmra_fragment_stored = on_dmra_fragment_stored
         self._config = config.get("SYSTEMS", {}).get(system_name, {})
         if self._config.get("MODE") == "OPENBRIDGE":
             self._laststrid = deque([], 20)
@@ -177,10 +185,14 @@ class HBPProtocol(DatagramProtocol):
             self._peers = self._config.setdefault("PEERS", {})
             self._dmra_by_stream: dict[bytes, dict[str, Any]] = {}
             self._dmra_rf_stream: dict[tuple[bytes, bytes], bytes] = {}
+            self._ta_voice_acc: dict[bytes, dict[int, Any]] = {}
+            self._ta_decoded_logged: set[bytes] = set()
         else:
             self._peers = {}
             self._dmra_by_stream = {}
             self._dmra_rf_stream = {}
+            self._ta_voice_acc = {}
+            self._ta_decoded_logged = set()
         if self._config.get("MODE") == "PEER":
             self._dmra_downlink: dict[bytes, dict[str, Any]] = {}
         if self._config.get("MODE") == "PEER":
@@ -250,8 +262,6 @@ class HBPProtocol(DatagramProtocol):
         if not parsed:
             return
         rf_src, block_id, payload = parsed
-        if block_id > 3:
-            return
         stream_id = self._dmra_rf_stream.get((peer_id, rf_src))
         if not stream_id:
             return
@@ -260,9 +270,62 @@ class HBPProtocol(DatagramProtocol):
             stream_id,
             {"blocks": {}, "rf_src": rf_src, "peer": peer_id, "last": now},
         )
-        entry["blocks"][block_id] = payload
+        if not store_ta_block(entry["blocks"], block_id, payload):
+            return
         entry["last"] = now
         entry["rf_src"] = rf_src
+        if self._on_dmra_fragment_stored:
+            self._on_dmra_fragment_stored(self._system, peer_id, rf_src, stream_id)
+
+    def store_ta_from_voice_burst(
+        self,
+        peer_id: bytes,
+        rf_src: bytes,
+        stream_id: bytes,
+        vseq: int,
+        dmrpkt: bytes,
+    ) -> None:
+        """Buffer TA from embedded LC in voice bursts B–E (MMDVM DMRSlot path)."""
+        if self._config.get("MODE") != "MASTER" or not stream_id:
+            return
+        if not self._CONFIG.get("GLOBAL", {}).get("TALKER_ALIAS", False):
+            return
+        entry = self._dmra_by_stream.setdefault(
+            stream_id,
+            {"blocks": {}, "rf_src": rf_src, "peer": peer_id, "last": time.time()},
+        )
+        acc = self._ta_voice_acc.setdefault(stream_id, {})
+        if try_buffer_ta_from_voice_fragments(acc, vseq, dmrpkt, entry["blocks"]):
+            entry["last"] = time.time()
+            entry["rf_src"] = rf_src
+            entry["peer"] = peer_id
+            if stream_id not in self._ta_decoded_logged:
+                text = decode_ta_from_blocks(entry["blocks"])
+                if text:
+                    self._ta_decoded_logged.add(stream_id)
+                    logger.debug(
+                        "(%s) *TALKER ALIAS* decoded '%s' from embedded voice (src %s stream %s)",
+                        self._system, text, int_id(rf_src), int_id(stream_id),
+                    )
+            if self._on_dmra_fragment_stored:
+                self._on_dmra_fragment_stored(self._system, peer_id, rf_src, stream_id)
+
+    def clear_ta_stream_buffer(self, stream_id: bytes) -> None:
+        self._dmra_by_stream.pop(stream_id, None)
+        self._ta_voice_acc.pop(stream_id, None)
+        self._ta_decoded_logged.discard(stream_id)
+
+    def copy_ta_stream_buffer(self, from_stream: bytes, to_stream: bytes) -> None:
+        """Carry decoded TA blocks from recording stream to parrot playback stream."""
+        entry = self._dmra_by_stream.get(from_stream)
+        if not entry or not entry.get("blocks"):
+            return
+        self._dmra_by_stream[to_stream] = {
+            "blocks": dict(entry["blocks"]),
+            "rf_src": entry.get("rf_src", b""),
+            "peer": entry.get("peer", b""),
+            "last": time.time(),
+        }
 
     def get_dmra_blocks(self, stream_id: bytes) -> dict[int, bytes] | None:
         """Return buffered DMRA block payloads for a stream, if any."""
@@ -281,6 +344,8 @@ class HBPProtocol(DatagramProtocol):
         for stream_id in list(self._dmra_by_stream):
             if self._dmra_by_stream[stream_id].get("last", 0) < cutoff:
                 del self._dmra_by_stream[stream_id]
+                self._ta_voice_acc.pop(stream_id, None)
+                self._ta_decoded_logged.discard(stream_id)
 
     def send_dmra_to_peers(self, packets: list[bytes], exclude_peer: bytes | None = None) -> int:
         """Send DMRA packets to logged-in peers (MASTER downlink). Returns peer count."""
@@ -567,6 +632,15 @@ class HBPProtocol(DatagramProtocol):
                 if sub_map is not None:
                     sub_map[_rf_src] = (self._system, _slot, pkt_time)
                 self.note_dmrd_stream(_peer_id, _rf_src, _stream_id)
+                if (
+                    _call_type in ("group", "vcsbk")
+                    and _frame_type != HBPF_DATA_SYNC
+                    and _dtype_vseq in (1, 2, 3, 4)
+                    and len(_data) >= 53
+                ):
+                    self.store_ta_from_voice_burst(
+                        _peer_id, _rf_src, _stream_id, _dtype_vseq, _data[20:53],
+                    )
                 if self._config.get("REPEAT", True):
                     pkt = [_data[:11], b"", _data[15:]]
                     for _peer in self._peers:
@@ -838,8 +912,16 @@ class HBPProtocol(DatagramProtocol):
                     _peer_id = pid
                     break
             if _peer_id is not None and len(_data) >= DMRA_PACKET_LEN:
-                logger.debug("(%s) Peer has sent Talker Alias packet %s", self._system, _data)
-                self.store_dmra_packet(_peer_id, _data)
+                if parse_dmra_packet(_data):
+                    logger.debug("(%s) Peer has sent Talker Alias packet %s", self._system, _data)
+                    self.store_dmra_packet(_peer_id, _data)
+                else:
+                    logger.debug(
+                        "(%s) Peer DMRA ignored (MMDVM expects byte7=0-3, got %s); raw %s",
+                        self._system,
+                        _data[7],
+                        _data,
+                    )
 
         elif _command == PRIN:
             logger.info("(%s) *ProxyInfo* Connection from IP:Port: %s", self._system, _data.decode("utf8", errors="replace")[4:])
@@ -1575,6 +1657,7 @@ def HBPProtocolFactory(
     on_obp_bcsq_received: Callable[[str, bytes, bytes], None] | None = None,
     on_talker_alias_local_repeat: Callable[[str, bytes, bytes, bytes], None] | None = None,
     on_talker_alias_stream_end: Callable[[str, bytes], None] | None = None,
+    on_dmra_fragment_stored: Callable[[str, bytes, bytes, bytes], None] | None = None,
 ) -> HBPProtocol:
     """Create HBP protocol instance (legacy: one HBSYSTEM per system)."""
     return HBPProtocol(
@@ -1592,4 +1675,5 @@ def HBPProtocolFactory(
         on_obp_bcsq_received=on_obp_bcsq_received,
         on_talker_alias_local_repeat=on_talker_alias_local_repeat,
         on_talker_alias_stream_end=on_talker_alias_stream_end,
+        on_dmra_fragment_stored=on_dmra_fragment_stored,
     )
diff --git a/src/adn_server/main.py b/src/adn_server/main.py
index 4d45b83..1fcbafb 100644
--- a/src/adn_server/main.py
+++ b/src/adn_server/main.py
@@ -322,6 +322,7 @@ def main() -> None:
         send_bcsq=send_bcsq,
         send_dmra_to_system=send_dmra_to_system,
         get_dmra_blocks=get_dmra_blocks,
+        call_later=reactor.callLater,
     )
     bridge_use_cases.apply_startup_bridges()
     report_factory.set_bridges(bridge_router.get_bridges())
@@ -481,6 +482,7 @@ def main() -> None:
             on_obp_bcsq_received=bridge_use_cases.on_obp_bcsq_received,
             on_talker_alias_local_repeat=bridge_use_cases.send_talker_alias_local_repeat,
             on_talker_alias_stream_end=bridge_use_cases.clear_talker_alias_stream,
+            on_dmra_fragment_stored=bridge_use_cases.on_dmra_fragment_stored,
         )
 
     def _listen_system(_name: str, bind: BindSpec, protocol: Any) -> Any: