Add multilingual STT support and config UI/runtime updates

tests/test_aman.py

@@ -86,11 +86,33 @@ class FakeHintModel:
         return [FakeSegment(self.text)], self.last_kwargs
 
 
+class FakeKwargModel:
+    def __init__(self, text: str = "hello world"):
+        self.text = text
+        self.last_kwargs = {}
+
+    def transcribe(self, _audio, **kwargs):
+        self.last_kwargs = dict(kwargs)
+        return [FakeSegment(self.text)], self.last_kwargs
+
+
+class FakeUnsupportedLanguageModel:
+    def __init__(self, text: str = "hello world"):
+        self.text = text
+        self.calls = []
+
+    def transcribe(self, _audio, language=None, vad_filter=None):
+        self.calls.append({"language": language, "vad_filter": vad_filter})
+        if language:
+            raise RuntimeError(f"unsupported language: {language}")
+        return [FakeSegment(self.text)], {"language": language, "vad_filter": vad_filter}
+
+
 class FakeAIProcessor:
     def __init__(self):
         self.last_kwargs = {}
 
-    def process(self, text, lang="en", **_kwargs):
+    def process(self, text, lang="auto", **_kwargs):
         self.last_kwargs = {"lang": lang, **_kwargs}
         return text
 
@@ -198,9 +220,10 @@ class DaemonTests(unittest.TestCase):
 
         daemon = self._build_daemon(desktop, model, cfg=cfg, verbose=False)
 
-        result = daemon._transcribe(object())
+        result, used_lang = daemon._transcribe(object())
 
         self.assertEqual(result, "hello")
+        self.assertEqual(used_lang, "auto")
         self.assertNotIn("hotwords", model.last_kwargs)
         self.assertNotIn("initial_prompt", model.last_kwargs)
 
@@ -213,13 +236,60 @@ class DaemonTests(unittest.TestCase):
 
         daemon = self._build_daemon(desktop, model, cfg=cfg, verbose=False)
 
-        result = daemon._transcribe(object())
+        result, used_lang = daemon._transcribe(object())
 
         self.assertEqual(result, "hello")
+        self.assertEqual(used_lang, "auto")
         self.assertIn("Docker", model.last_kwargs["hotwords"])
         self.assertIn("Systemd", model.last_kwargs["hotwords"])
         self.assertIn("Preferred vocabulary", model.last_kwargs["initial_prompt"])
 
+    def test_transcribe_uses_configured_language_hint(self):
+        desktop = FakeDesktop()
+        model = FakeModel(text="hola")
+        cfg = self._config()
+        cfg.stt.language = "es"
+
+        daemon = self._build_daemon(desktop, model, cfg=cfg, verbose=False)
+
+        result, used_lang = daemon._transcribe(object())
+
+        self.assertEqual(result, "hola")
+        self.assertEqual(used_lang, "es")
+        self.assertEqual(model.last_kwargs["language"], "es")
+
+    def test_transcribe_auto_language_omits_language_kwarg(self):
+        desktop = FakeDesktop()
+        model = FakeKwargModel(text="hello")
+        cfg = self._config()
+        cfg.stt.language = "auto"
+
+        daemon = self._build_daemon(desktop, model, cfg=cfg, verbose=False)
+
+        result, used_lang = daemon._transcribe(object())
+
+        self.assertEqual(result, "hello")
+        self.assertEqual(used_lang, "auto")
+        self.assertNotIn("language", model.last_kwargs)
+
+    def test_transcribe_falls_back_to_auto_when_hint_is_rejected(self):
+        desktop = FakeDesktop()
+        model = FakeUnsupportedLanguageModel(text="bonjour")
+        cfg = self._config()
+        cfg.stt.language = "fr"
+
+        daemon = self._build_daemon(desktop, model, cfg=cfg, verbose=False)
+
+        with self.assertLogs(level="WARNING") as logs:
+            result, used_lang = daemon._transcribe(object())
+
+        self.assertEqual(result, "bonjour")
+        self.assertEqual(used_lang, "auto")
+        self.assertEqual(len(model.calls), 2)
+        self.assertEqual(model.calls[0]["language"], "fr")
+        self.assertIsNone(model.calls[1]["language"])
+        self.assertTrue(any("falling back to auto-detect" in line for line in logs.output))
+
     def test_verbose_flag_controls_transcript_logging(self):
         desktop = FakeDesktop()
         cfg = self._config()
@@ -237,7 +307,7 @@ class DaemonTests(unittest.TestCase):
         ) as processor_cls:
             daemon = aman.Daemon(self._config(), desktop, verbose=True)
 
-        processor_cls.assert_called_once_with(verbose=True)
+        processor_cls.assert_called_once_with(verbose=True, model_path=None)
         self.assertIsNotNone(daemon.ai_processor)
 
     @patch("aman.stop_audio_recording", return_value=FakeAudio(8))
@@ -333,6 +403,31 @@ class DaemonTests(unittest.TestCase):
 
         self.assertEqual(ai_processor.last_kwargs.get("profile"), "fast")
 
+    @patch("aman.stop_audio_recording", return_value=FakeAudio(8))
+    @patch("aman.start_audio_recording", return_value=(object(), object()))
+    def test_ai_processor_receives_effective_language(self, _start_mock, _stop_mock):
+        desktop = FakeDesktop()
+        cfg = self._config()
+        cfg.stt.language = "es"
+        ai_processor = FakeAIProcessor()
+        daemon = self._build_daemon(
+            desktop,
+            FakeModel(text="hola mundo"),
+            cfg=cfg,
+            verbose=False,
+            ai_processor=ai_processor,
+        )
+        daemon._start_stop_worker = (
+            lambda stream, record, trigger, process_audio: daemon._stop_and_process(
+                stream, record, trigger, process_audio
+            )
+        )
+
+        daemon.toggle()
+        daemon.toggle()
+
+        self.assertEqual(ai_processor.last_kwargs.get("lang"), "es")
+
     @patch("aman.start_audio_recording")
     def test_paused_state_blocks_recording_start(self, start_mock):
         desktop = FakeDesktop()