feat: add OpenAI-compatible Chatterbox TTS service

tests/test_synthesis.py

@@ -0,0 +1,51 @@
+"""CUDA smoke coverage for the real Chatterbox synthesis path."""
+
+from pathlib import Path
+
+import pytest
+import torch
+import torchaudio
+
+import app as chatterbox_app
+
+
+def test_runtime_requirements_pin_chatterbox_cuda_contract():
+    requirements = {
+        line.strip()
+        for line in (Path(__file__).parents[1] / "requirements.txt").read_text().splitlines()
+        if line.strip() and not line.startswith("#")
+    }
+
+    assert {
+        "chatterbox-tts==0.1.7",
+        "torch==2.6.0",
+        "torchaudio==2.6.0",
+        "pydantic",
+        "setuptools<81",
+        "pytest",
+    } <= requirements
+
+
+def test_chatterbox_generates_cuda_audio_from_reference_voice(tmp_path: Path):
+    """The app wrapper returns real, decodable CUDA-synthesized WAV audio."""
+    if not torch.cuda.is_available():
+        pytest.skip("CUDA is unavailable")
+
+    reference = chatterbox_app.VOICES_DIR / "reference.wav"
+    assert reference.is_file(), "voices/reference.wav must be available for synthesis"
+    manager = chatterbox_app.model_manager
+
+    try:
+        wav_bytes = manager.generate("Chatterbox CUDA smoke test.", reference, 1.0)
+    finally:
+        manager.unload()
+
+    output_path = tmp_path / "chatterbox-cuda-smoke.wav"
+    output_path.write_bytes(wav_bytes)
+    waveform, sample_rate = torchaudio.load(output_path)
+    assert isinstance(waveform, torch.Tensor)
+    assert waveform.ndim == 2
+    assert waveform.shape[0] == 1
+    assert waveform.numel() > 0
+    assert sample_rate > 0
+    assert output_path.stat().st_size > 44