diff --git a/surfsense_backend/tests/unit/podcasts/test_voice_catalog.py b/surfsense_backend/tests/unit/podcasts/test_voice_catalog.py
new file mode 100644
index 000000000..d94c85922
--- /dev/null
+++ b/surfsense_backend/tests/unit/podcasts/test_voice_catalog.py
@@ -0,0 +1,103 @@
+"""The voice catalog and provider identification.
+
+The catalog is the single source of truth for which voices exist; resolution,
+the API picker, and the renderer all depend on its lookups behaving correctly.
+These tests build a small catalog of their own so they assert on the lookup
+behavior, not on which specific voices ship.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from app.podcasts.voices import (
+    ANY_LANGUAGE,
+    CatalogVoice,
+    TtsProvider,
+    VoiceCatalog,
+    VoiceGender,
+    provider_from_service,
+)
+
+pytestmark = pytest.mark.unit
+
+
+def _voice(
+    voice_id: str,
+    *,
+    provider: TtsProvider = TtsProvider.KOKORO,
+    language: str = "en-US",
+    gender: VoiceGender = VoiceGender.MALE,
+) -> CatalogVoice:
+    return CatalogVoice(
+        voice_id=voice_id,
+        provider=provider,
+        language=language,
+        display_name=voice_id,
+        gender=gender,
+        native_ref=voice_id,
+    )
+
+
+def test_for_provider_returns_only_that_providers_voices():
+    catalog = VoiceCatalog(
+        [
+            _voice("k1", provider=TtsProvider.KOKORO),
+            _voice("o1", provider=TtsProvider.OPENAI),
+        ]
+    )
+    assert [v.voice_id for v in catalog.for_provider(TtsProvider.KOKORO)] == ["k1"]
+
+
+def test_for_language_matches_on_the_primary_subtag():
+    """A request for 'en' should match an 'en-US' voice (region-insensitive)."""
+    catalog = VoiceCatalog([_voice("k1", language="en-US")])
+    assert [v.voice_id for v in catalog.for_language(TtsProvider.KOKORO, "en")] == ["k1"]
+
+
+def test_for_language_excludes_other_languages():
+    catalog = VoiceCatalog([_voice("k1", language="en-US")])
+    assert catalog.for_language(TtsProvider.KOKORO, "fr") == []
+
+
+def test_an_any_language_voice_speaks_every_language():
+    """Provider-agnostic voices (e.g. OpenAI) match whatever the text is in."""
+    voice = _voice("o1", provider=TtsProvider.OPENAI, language=ANY_LANGUAGE)
+    assert voice.speaks("ja")
+    assert voice.speaks("pt-BR")
+
+
+def test_supports_language_reports_availability():
+    catalog = VoiceCatalog([_voice("k1", language="en-US")])
+    assert catalog.supports_language(TtsProvider.KOKORO, "en")
+    assert not catalog.supports_language(TtsProvider.KOKORO, "de")
+
+
+def test_get_raises_for_an_unknown_voice():
+    catalog = VoiceCatalog([_voice("k1")])
+    with pytest.raises(KeyError):
+        catalog.get("nope")
+
+
+def test_a_catalog_rejects_duplicate_voice_ids():
+    """Stored ids must be unique so a brief's voice_id resolves unambiguously."""
+    with pytest.raises(ValueError):
+        VoiceCatalog([_voice("dup"), _voice("dup")])
+
+
+@pytest.mark.parametrize(
+    ("service", "expected"),
+    [
+        ("openai/tts-1", TtsProvider.OPENAI),
+        ("azure/neural", TtsProvider.AZURE),
+        ("vertex_ai/some-model", TtsProvider.VERTEX_AI),
+        ("local/kokoro", TtsProvider.KOKORO),
+    ],
+)
+def test_provider_is_identified_from_the_config_string(service, expected):
+    assert provider_from_service(service) == expected
+
+
+def test_unknown_provider_prefix_is_rejected():
+    with pytest.raises(ValueError):
+        provider_from_service("madeup/model")