Function generate_synthetic_stem

ci/tools/generate_audio_fixtures.py:160–330

Generate a synthetic audio stem using ffmpeg's signal generators.

(ffmpeg: str, stem_type: str, output: Path)

def generate_synthetic_stem(ffmpeg: str, stem_type: str, output: Path) -> None:
    """Generate a synthetic audio stem using ffmpeg's signal generators."""
    if stem_type == "voice_male":
        # Synthetic male voice: F0=130Hz with formants at F1=700Hz, F2=1200Hz
        # Uses multiple harmonics shaped by formant envelopes
        expr_parts: list[str] = []
        f0 = 130.0
        for h in range(1, 30):
            freq = f0 * h
            if freq > SAMPLE_RATE / 2:
                break
            # Natural 1/h rolloff
            amp = 1.0 / h
            # Gaussian formant shaping (F1=700, F2=1200, F3=2600)
            import math

            f1_gain = math.exp(-0.5 * ((freq - 700) / 150) ** 2)
            f2_gain = math.exp(-0.5 * ((freq - 1200) / 200) ** 2)
            f3_gain = math.exp(-0.5 * ((freq - 2600) / 250) ** 2)
            formant = max(f1_gain, f2_gain, f3_gain, 0.02)
            final_amp = amp * formant * 0.3
            expr_parts.append(f"{final_amp:.4f}*sin(2*PI*{freq:.1f}*t)")
        expr = "+".join(expr_parts)
        # Add slight amplitude modulation for naturalness
        expr = f"({expr})*( 0.8 + 0.2*sin(2*PI*5.5*t) )"
        run_ffmpeg(
            ffmpeg,
            [
                "-f",
                "lavfi",
                "-i",
                f"aevalsrc={expr}:s={SAMPLE_RATE}:d={DURATION_S}",
                "-ac",
                str(CHANNELS),
                "-c:a",
                "libvorbis",
                "-b:a",
                OGG_BITRATE,
                str(output),
            ],
            f"Synthesizing {stem_type}",
        )

    elif stem_type == "voice_female":
        # Synthetic female voice: F0=220Hz with formants at F1=800, F2=1500
        expr_parts: list[str] = []
        f0 = 220.0
        for h in range(1, 20):
            freq = f0 * h
            if freq > SAMPLE_RATE / 2:
                break
            amp = 1.0 / h
            import math

            f1_gain = math.exp(-0.5 * ((freq - 800) / 160) ** 2)
            f2_gain = math.exp(-0.5 * ((freq - 1500) / 220) ** 2)
            f3_gain = math.exp(-0.5 * ((freq - 2800) / 280) ** 2)
            formant = max(f1_gain, f2_gain, f3_gain, 0.02)