feat: support audio+text input for forced alignment and auto-sync server.py

- Support simultaneous audio + text input in infer() for models like
  fa-zh that require both (forced alignment). Server constructs tuple
  input with data_type=("sound", "text") automatically.
- Auto-sync server.py from template on start() by comparing file
  contents, so package upgrades take effect without re-install.

Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>

Author: 3 people

src/funasr_server/client.py

@@ -27,6 +27,7 @@
 import logging
 import os
 import platform
+import shutil
 import signal
 import subprocess
 import time
@@ -39,6 +40,8 @@
 from funasr_server.mirror import get_hub
 from funasr_server.models import resolve_model_id
 
+_TEMPLATE_DIR = Path(__file__).parent / "runtime_template"
+
 logger = logging.getLogger(__name__)
 
 
@@ -231,6 +234,8 @@ def start(self, timeout: float = 60) -> int:
             logger.info(f"Server already running (pid={self._process.pid})")
             return self.port
 
+        self._sync_server_py()
+
         uv_path = self.installer.get_uv_path()
         if not uv_path:
             raise RuntimeError("uv not found. Call ensure_installed() first.")
@@ -303,6 +308,22 @@ def is_running(self) -> bool:
             return False
         return self._process.poll() is None
 
+    def _sync_server_py(self):
+        """Sync server.py from template to runtime dir if outdated.
+
+        Called automatically before each start() so that upgrading the
+        funasr-server package immediately picks up server-side changes
+        without requiring a full re-install.
+        """
+        template = _TEMPLATE_DIR / "server.py"
+        runtime = self.runtime_dir / "server.py"
+        if not template.exists() or not runtime.exists():
+            return
+        if runtime.read_bytes() == template.read_bytes():
+            return
+        shutil.copy2(template, runtime)
+        logger.info("Synced server.py from template (updated)")
+
     # ------------------------------------------------------------------
     # High-level API
     # ------------------------------------------------------------------
@@ -423,11 +444,15 @@ def infer(
 
         if audio_bytes is not None:
             params["input_base64"] = base64.b64encode(audio_bytes).decode()
+            if text is not None:
+                params["text"] = text
         elif audio is not None:
             if os.path.exists(audio):
                 params["input"] = str(Path(audio).resolve())
             else:
                 params["input"] = audio
+            if text is not None:
+                params["text"] = text
         elif text is not None:
             params["input"] = text
         else:

src/funasr_server/runtime_template/server.py

@@ -170,8 +170,10 @@ def rpc_infer(params: dict) -> dict:
     input_base64 = params.get("input_base64")
     tmp_file = None
 
+    text = params.get("text")
+
     # Build generate kwargs: everything except control params
-    _control_keys = {"name", "input", "input_base64", "audio_format"}
+    _control_keys = {"name", "input", "input_base64", "audio_format", "text"}
     generate_kwargs = {k: v for k, v in params.items() if k not in _control_keys}
 
     if input_base64:
@@ -186,6 +188,12 @@ def rpc_infer(params: dict) -> dict:
     if input_data is None:
         raise ValueError("'input' or 'input_base64' is required")
 
+    # For models that need both audio and text (e.g. fa-zh forced alignment),
+    # pass input as a tuple with data_type hint.
+    if text is not None and input_data is not None:
+        input_data = (input_data, text)
+        generate_kwargs["data_type"] = ("sound", "text")
+
     def _on_progress(current, total):
         _progress[name] = {"current": current, "total": total}
 

tests/test_client.py

@@ -461,3 +461,55 @@ def test_model_infer_without_progress_callback(client):
     # Without progress_callback — should work exactly as before
     result = model.infer(audio="test.wav")
     assert result[0]["text"] == "hello world"
+
+
+# ------------------------------------------------------------------
+# server.py sync tests
+# ------------------------------------------------------------------
+
+def test_sync_server_py_copies_when_different(tmp_path):
+    """_sync_server_py copies template when runtime file differs."""
+    runtime = tmp_path / "runtime"
+    runtime.mkdir()
+    server_py = runtime / "server.py"
+    server_py.write_text("old content")
+
+    asr = FunASR(runtime_dir=str(runtime))
+
+    template = Path(__file__).parent.parent / "src" / "funasr_server" / "runtime_template" / "server.py"
+    template_content = template.read_bytes()
+
+    asr._sync_server_py()
+
+    assert server_py.read_bytes() == template_content
+
+
+def test_sync_server_py_skips_when_same(tmp_path):
+    """_sync_server_py does nothing when files are identical."""
+    runtime = tmp_path / "runtime"
+    runtime.mkdir()
+    server_py = runtime / "server.py"
+
+    template = Path(__file__).parent.parent / "src" / "funasr_server" / "runtime_template" / "server.py"
+    template_content = template.read_bytes()
+    server_py.write_bytes(template_content)
+
+    mtime_before = server_py.stat().st_mtime
+
+    asr = FunASR(runtime_dir=str(runtime))
+    asr._sync_server_py()
+
+    # File should not have been touched (same mtime)
+    assert server_py.stat().st_mtime == mtime_before
+
+
+def test_sync_server_py_skips_when_no_runtime(tmp_path):
+    """_sync_server_py does nothing when runtime server.py doesn't exist yet."""
+    runtime = tmp_path / "runtime"
+    runtime.mkdir()
+    # No server.py in runtime dir
+
+    asr = FunASR(runtime_dir=str(runtime))
+    asr._sync_server_py()  # should not raise
+
+    assert not (runtime / "server.py").exists()