From 0afaaf561b90afad5cca93effbb65f658a4d3ba1 Mon Sep 17 00:00:00 2001
From: JB <jb-apps@outlook.com>
Date: Mon, 15 Dec 2025 16:28:30 -0800
Subject: [PATCH] feat: Implement voice profile resolution in debug TTS and add
 corresponding tests

---
 abogen/web/debug_tts_runner.py  | 25 ++++++++++++++++++++++++
 tests/test_debug_tts_samples.py | 34 +++++++++++++++++++++++++++++++++
 2 files changed, 59 insertions(+)

diff --git a/abogen/web/debug_tts_runner.py b/abogen/web/debug_tts_runner.py
index d8996ef..04871e0 100644
--- a/abogen/web/debug_tts_runner.py
+++ b/abogen/web/debug_tts_runner.py
@@ -28,6 +28,18 @@ class DebugWavArtifact:
     code: Optional[str] = None
 
 
+def _resolve_voice_setting(value: str) -> tuple[str, Optional[str], Optional[str]]:
+    """Resolve settings voice strings into a pipeline-ready voice spec.
+
+    Supports "profile:<name>" by converting it into a concrete voice formula.
+    Returns (resolved_voice_spec, profile_name, profile_language).
+    """
+
+    from abogen.web.routes.utils.voice import resolve_voice_setting
+
+    return resolve_voice_setting(value)
+
+
 def _load_pipeline(language: str, use_gpu: bool) -> Any:
     device = "cpu"
     if use_gpu:
@@ -117,6 +129,19 @@ def run_debug_tts_wavs(
     use_gpu = bool(settings.get("use_gpu", False))
     speed = float(settings.get("default_speed", 1.0) or 1.0)
 
+    # Settings may store "profile:<name>" which is not a Kokoro voice ID.
+    # Resolve it to a concrete voice formula (e.g. "af_heart*0.5+...") so Kokoro
+    # doesn't attempt to download a non-existent "voices/profile:<name>.pt".
+    try:
+        resolved_voice, _profile_name, profile_language = _resolve_voice_setting(voice_spec)
+        if resolved_voice:
+            voice_spec = resolved_voice
+        if profile_language:
+            language = str(profile_language).strip() or language
+    except Exception:
+        # Voice profile resolution is best-effort; fall back to raw voice_spec.
+        pass
+
     # Best-effort voice caching (only for known Kokoro internal voices).
     voice_ids = _spec_to_voice_ids(voice_spec)
     if voice_ids:
diff --git a/tests/test_debug_tts_samples.py b/tests/test_debug_tts_samples.py
index 3b72b14..cf9c3b9 100644
--- a/tests/test_debug_tts_samples.py
+++ b/tests/test_debug_tts_samples.py
@@ -119,3 +119,37 @@ def test_debug_samples_have_minimum_per_category():
 
     for prefix, minimum in prefixes.items():
         assert counts[prefix] >= minimum
+
+
+def test_debug_runner_resolves_profile_voice_before_pipeline(tmp_path, monkeypatch):
+    from abogen.web import debug_tts_runner as runner
+
+    # Stub voice setting resolution so we don't depend on the user's profile file.
+    monkeypatch.setattr(runner, "_resolve_voice_setting", lambda value: ("af_heart", "AM HQ Alt", None))
+
+    calls = []
+
+    class _Seg:
+        def __init__(self, audio):
+            self.audio = audio
+
+    class DummyPipeline:
+        def __call__(self, text, **kwargs):
+            calls.append(kwargs.get("voice"))
+            audio = np.zeros(int(0.05 * runner.SAMPLE_RATE), dtype="float32")
+            yield _Seg(audio)
+
+    monkeypatch.setattr(runner, "_load_pipeline", lambda language, use_gpu: DummyPipeline())
+
+    settings = {
+        "language": "en",
+        "default_voice": "profile:AM HQ Alt",
+        "use_gpu": False,
+        "default_speed": 1.0,
+    }
+
+    manifest = runner.run_debug_tts_wavs(output_root=tmp_path, settings=settings)
+    assert manifest.get("run_id")
+    assert calls
+    # Must not pass through the profile:* string.
+    assert all(isinstance(v, str) and not v.lower().startswith("profile:") for v in calls)