diff --git a/tests/test_transcriber.py b/tests/test_transcriber.py
new file mode 100644
index 0000000..d43c434
--- /dev/null
+++ b/tests/test_transcriber.py
@@ -0,0 +1,57 @@
+from unittest.mock import MagicMock, patch
+
+import numpy as np
+import pytest
+
+from whisper_local.transcriber import Transcriber
+
+
+class TestTranscriber:
+    @patch("whisper_local.transcriber.WhisperModel")
+    def test_init_loads_model(self, mock_model_class):
+        t = Transcriber(model_name="small", compute_type="int8", language="de")
+        mock_model_class.assert_called_once_with("small", compute_type="int8")
+        assert t.language == "de"
+
+    @patch("whisper_local.transcriber.WhisperModel")
+    def test_transcribe_returns_text(self, mock_model_class):
+        mock_model = MagicMock()
+        mock_segment = MagicMock()
+        mock_segment.text = " Hallo Welt "
+        mock_model.transcribe.return_value = ([mock_segment], None)
+        mock_model_class.return_value = mock_model
+
+        t = Transcriber(model_name="small", compute_type="int8", language="de")
+        audio = np.zeros(16000, dtype=np.float32)
+        result = t.transcribe(audio)
+
+        assert result == "Hallo Welt"
+        mock_model.transcribe.assert_called_once_with(audio, language="de")
+
+    @patch("whisper_local.transcriber.WhisperModel")
+    def test_transcribe_empty_segments(self, mock_model_class):
+        mock_model = MagicMock()
+        mock_model.transcribe.return_value = ([], None)
+        mock_model_class.return_value = mock_model
+
+        t = Transcriber(model_name="small", compute_type="int8", language="de")
+        audio = np.zeros(16000, dtype=np.float32)
+        result = t.transcribe(audio)
+
+        assert result == ""
+
+    @patch("whisper_local.transcriber.WhisperModel")
+    def test_transcribe_multiple_segments(self, mock_model_class):
+        mock_model = MagicMock()
+        seg1 = MagicMock()
+        seg1.text = " Hallo "
+        seg2 = MagicMock()
+        seg2.text = " Welt "
+        mock_model.transcribe.return_value = ([seg1, seg2], None)
+        mock_model_class.return_value = mock_model
+
+        t = Transcriber(model_name="small", compute_type="int8", language="de")
+        audio = np.zeros(16000, dtype=np.float32)
+        result = t.transcribe(audio)
+
+        assert result == "Hallo Welt"
diff --git a/whisper_local/transcriber.py b/whisper_local/transcriber.py
new file mode 100644
index 0000000..614a13b
--- /dev/null
+++ b/whisper_local/transcriber.py
@@ -0,0 +1,26 @@
+"""Whisper-Transkription via faster-whisper."""
+
+import logging
+
+import numpy as np
+from faster_whisper import WhisperModel
+
+logger = logging.getLogger(__name__)
+
+
+class Transcriber:
+    def __init__(self, model_name: str = "small", compute_type: str = "int8", language: str = "de"):
+        self.language = language
+        logger.info("Lade Whisper-Modell '%s' (compute_type=%s)...", model_name, compute_type)
+        self.model = WhisperModel(model_name, compute_type=compute_type)
+        logger.info("Modell geladen")
+
+    def transcribe(self, audio: np.ndarray) -> str:
+        """Transkribiert Audio-Array zu Text."""
+        segments, _ = self.model.transcribe(audio, language=self.language)
+        text = " ".join(segment.text.strip() for segment in segments if segment.text.strip())
+        if text:
+            logger.info("Transkribiert: %s", text)
+        else:
+            logger.info("Keine Sprache erkannt")
+        return text