feat: Enhance voice formula parsing and validation, implement voice asset caching, and add tests for new functionality

tests/test_prepare_form.py

@@ -0,0 +1,61 @@
+from pathlib import Path
+
+from werkzeug.datastructures import MultiDict
+
+from abogen.web.routes import _apply_prepare_form
+from abogen.web.service import PendingJob
+
+
+def _make_pending_job() -> PendingJob:
+    return PendingJob(
+        id="pending",
+        original_filename="example.epub",
+        stored_path=Path("example.epub"),
+        language="a",
+        voice="af_nova",
+        speed=1.0,
+        use_gpu=False,
+        subtitle_mode="none",
+        output_format="mp3",
+        save_mode="save_next_to_input",
+        output_folder=None,
+        replace_single_newlines=False,
+        subtitle_format="srt",
+        total_characters=0,
+        save_chapters_separately=False,
+        merge_chapters_at_end=True,
+        separate_chapters_format="wav",
+        silence_between_chapters=2.0,
+        save_as_project=False,
+        voice_profile=None,
+        max_subtitle_words=50,
+        metadata_tags={},
+        chapters=[],
+        created_at=0.0,
+    )
+
+
+def test_apply_prepare_form_handles_custom_mix_for_speakers():
+    pending = _make_pending_job()
+    pending.speakers = {
+        "hero": {
+            "id": "hero",
+            "label": "Hero",
+        }
+    }
+
+    form = MultiDict(
+        {
+            "chapter_intro_delay": "0.5",
+            "speaker-hero-voice": "__custom_mix",
+            "speaker-hero-formula": "af_nova*0.6+am_liam*0.4",
+        }
+    )
+
+    _, _, _, errors, *_ = _apply_prepare_form(pending, form)
+
+    assert not errors
+    hero = pending.speakers["hero"]
+    assert hero["voice_formula"] == "af_nova*0.6+am_liam*0.4"
+    assert hero["resolved_voice"] == "af_nova*0.6+am_liam*0.4"
+    assert "voice" not in hero or hero["voice"] != "__custom_mix"

tests/test_voice_cache.py

@@ -0,0 +1,56 @@
+from types import SimpleNamespace
+from typing import cast
+
+import pytest
+
+from abogen.constants import VOICES_INTERNAL
+from abogen.voice_cache import _CACHED_VOICES, ensure_voice_assets
+from abogen.web.conversion_runner import _collect_required_voice_ids
+from abogen.web.service import Job
+
+
+@pytest.fixture(autouse=True)
+def clear_voice_cache():
+    _CACHED_VOICES.clear()
+    yield
+    _CACHED_VOICES.clear()
+
+
+def test_ensure_voice_assets_downloads_missing(monkeypatch):
+    recorded = []
+
+    def fake_download(**kwargs):
+        recorded.append(kwargs["filename"])
+        return "/tmp/fake"
+
+    monkeypatch.setattr("abogen.voice_cache.hf_hub_download", fake_download)
+
+    downloaded, errors = ensure_voice_assets(["af_nova", "am_liam"])
+
+    assert downloaded == {"af_nova", "am_liam"}
+    assert errors == {}
+    assert recorded == ["voices/af_nova.pt", "voices/am_liam.pt"]
+
+    recorded.clear()
+    downloaded_again, errors_again = ensure_voice_assets(["af_nova"])
+
+    assert downloaded_again == set()
+    assert errors_again == {}
+    assert recorded == []
+
+
+def test_collect_required_voice_ids_includes_all():
+    job = SimpleNamespace(
+        voice="af_nova",
+        chapters=[{"voice_formula": "af_nova*0.7+am_liam*0.3"}],
+        chunks=[{"voice": "am_michael"}],
+        speakers={
+            "hero": {"voice_formula": "af_nova*0.6+am_liam*0.4"},
+            "narrator": {"voice": "af_nova"},
+        },
+    )
+
+    voices = _collect_required_voice_ids(cast(Job, job))
+
+    assert {"af_nova", "am_liam", "am_michael"}.issubset(voices)
+    assert voices.issuperset(VOICES_INTERNAL)