tests/test_omnivoice_adapter.py

from pathlib import Path

import numpy as np

from backend.omnivoice_adapter import OmniVoiceAdapter
from backend.types import VoiceConfig


class FakeModel:
    def __init__(self) -> None:
        self.generate_calls = []
        self.clone_prompt_calls = []

    def create_voice_clone_prompt(self, *, ref_audio, ref_text=None):
        self.clone_prompt_calls.append({"ref_audio": ref_audio, "ref_text": ref_text})
        return {"prompt": "cached"}

    def generate(self, **kwargs):
        self.generate_calls.append(kwargs)
        return [np.zeros(24000, dtype=np.float32)]


def test_omnivoice_adapter_uses_stable_generation_config_for_designed_voice(tmp_path: Path) -> None:
    adapter = OmniVoiceAdapter()
    fake_model = FakeModel()
    adapter._model = fake_model
    adapter._backend = "omnivoice"

    adapter.synthesize(
        text="Chapter one text.",
        output_path=tmp_path / "chapter.wav",
        voice_config=VoiceConfig(mode="design", design_prompt="male, elderly, low pitch, british accent"),
        diffusion_steps=32,
        speed=1.0,
    )

    generation_config = fake_model.generate_calls[0]["generation_config"]
    assert generation_config.position_temperature == 0.0
    assert generation_config.class_temperature == 0.0


def test_omnivoice_adapter_reuses_cached_clone_prompt_across_chapters(tmp_path: Path) -> None:
    adapter = OmniVoiceAdapter()
    fake_model = FakeModel()
    adapter._model = fake_model
    adapter._backend = "omnivoice"
    voice = VoiceConfig(mode="clone", sample_path="/tmp/sample.wav", reference_text="Reference speech")

    adapter.synthesize(
        text="Chapter one text.",
        output_path=tmp_path / "chapter-1.wav",
        voice_config=voice,
        diffusion_steps=32,
        speed=1.0,
    )
    adapter.synthesize(
        text="Chapter two text.",
        output_path=tmp_path / "chapter-2.wav",
        voice_config=voice,
        diffusion_steps=32,
        speed=1.0,
    )

    assert len(fake_model.clone_prompt_calls) == 1
    assert fake_model.generate_calls[0]["voice_clone_prompt"] == {"prompt": "cached"}
    assert fake_model.generate_calls[1]["voice_clone_prompt"] == {"prompt": "cached"}
    assert "ref_audio" not in fake_model.generate_calls[0]
    assert "ref_audio" not in fake_model.generate_calls[1]