feat: Implement unsupported character handling in Supertonic pipeline and add related tests

tests/test_manual_overrides_applied_first.py

@@ -0,0 +1,51 @@
+from abogen.web import conversion_runner
+
+
+class DummyJob:
+    def __init__(self):
+        self.language = "en"
+        self.voice = "M1"
+        self.speakers = None
+        self.manual_overrides = []
+        self.pronunciation_overrides = []
+
+
+def _apply(text: str, job: DummyJob) -> str:
+    merged = conversion_runner._merge_pronunciation_overrides(job)
+    rules = conversion_runner._compile_pronunciation_rules(merged)
+    return conversion_runner._apply_pronunciation_rules(text, rules)
+
+
+def test_manual_override_is_applied_even_if_pronunciation_overrides_stale():
+    job = DummyJob()
+    job.manual_overrides = [
+        {
+            "token": "Unfu*k",
+            "pronunciation": "Unfuck",
+        }
+    ]
+
+    out = _apply("He said Unfu*k loudly.", job)
+    assert "Unfuck" in out
+    assert "Unfu*k" not in out
+
+
+def test_manual_override_takes_precedence_over_existing_pronunciation_override():
+    job = DummyJob()
+    job.pronunciation_overrides = [
+        {
+            "token": "Unfu*k",
+            "normalized": "unfu*k",
+            "pronunciation": "WRONG",
+        }
+    ]
+    job.manual_overrides = [
+        {
+            "token": "Unfu*k",
+            "pronunciation": "RIGHT",
+        }
+    ]
+
+    out = _apply("Unfu*k.", job)
+    assert "RIGHT" in out
+    assert "WRONG" not in out

tests/test_tts_supertonic_unsupported_chars.py

@@ -0,0 +1,53 @@
+import numpy as np
+
+from abogen.tts_supertonic import SupertonicPipeline
+
+
+class _DummyTTS:
+    def get_voice_style(self, voice_name: str):
+        return {"voice": voice_name}
+
+    def synthesize(
+        self,
+        *,
+        text: str,
+        voice_style,
+        total_steps: int,
+        speed: float,
+        max_chunk_length: int,
+        silence_duration: float,
+        verbose: bool,
+    ):
+        if "•" in text:
+            raise ValueError("Found 1 unsupported character(s): ['•']")
+        # Return 50ms of audio at 24kHz.
+        sr = 24000
+        audio = np.zeros(int(0.05 * sr), dtype="float32")
+        return audio, 0.05
+
+
+def test_supertonic_pipeline_strips_unsupported_characters_and_retries():
+    # Avoid importing/initializing real supertonic by manually constructing the pipeline.
+    pipeline = SupertonicPipeline.__new__(SupertonicPipeline)
+    pipeline.sample_rate = 24000
+    pipeline.total_steps = 5
+    pipeline.max_chunk_length = 1000
+    pipeline._tts = _DummyTTS()
+
+    segs = list(pipeline("Hello • world", voice="M1", speed=1.0))
+    assert len(segs) == 1
+    assert segs[0].graphemes == "Hello  world" or segs[0].graphemes == "Hello world"
+    assert isinstance(segs[0].audio, np.ndarray)
+    assert segs[0].audio.dtype == np.float32
+    assert segs[0].audio.size > 0
+
+
+def test_supertonic_pipeline_drops_chunk_if_only_unsupported_characters():
+    pipeline = SupertonicPipeline.__new__(SupertonicPipeline)
+    pipeline.sample_rate = 24000
+    pipeline.total_steps = 5
+    pipeline.max_chunk_length = 1000
+    pipeline._tts = _DummyTTS()
+
+    segs = list(pipeline("•", voice="M1", speed=1.0))
+    assert segs == []