| from __future__ import annotations |
|
|
| import numpy as np |
|
|
| try: |
| from scipy.signal import butter, lfilter |
| except Exception: |
| butter = None |
| lfilter = None |
|
|
| from .audio_utils import ensure_2d_audio, normalize_audio |
|
|
|
|
| def _lowpass(audio: np.ndarray, sr: int, cutoff: float) -> np.ndarray: |
| if butter is None or lfilter is None: |
| kernel = np.ones(7, dtype=np.float32) / 7 |
| return np.apply_along_axis(lambda x: np.convolve(x, kernel, mode="same"), 0, audio).astype(np.float32) |
| b, a = butter(2, min(cutoff / (sr / 2), 0.98), btype="low") |
| return lfilter(b, a, audio, axis=0).astype(np.float32) |
|
|
|
|
| def _bandpass(audio: np.ndarray, sr: int, low: float, high: float) -> np.ndarray: |
| if butter is None or lfilter is None: |
| blurred = _lowpass(audio, sr, high) |
| very_low = _lowpass(audio, sr, low) |
| return (blurred - very_low * 0.35).astype(np.float32) |
| b, a = butter(2, [low / (sr / 2), min(high / (sr / 2), 0.98)], btype="band") |
| return lfilter(b, a, audio, axis=0).astype(np.float32) |
|
|
|
|
| def _compress(audio: np.ndarray, drive: float = 1.4) -> np.ndarray: |
| return np.tanh(audio * drive) / np.tanh(drive) |
|
|
|
|
| def _echo(audio: np.ndarray, sr: int, delay: float = 0.055, amount: float = 0.16) -> np.ndarray: |
| out = audio.copy() |
| offset = int(delay * sr) |
| if offset < len(out): |
| out[offset:] += audio[:-offset] * amount |
| return out |
|
|
|
|
| def apply_texture(audio: np.ndarray, sr: int, texture_id: str) -> np.ndarray: |
| audio = ensure_2d_audio(audio) |
| rng = np.random.default_rng(777) |
| texture_id = texture_id or "clean_digital" |
| if texture_id == "vinyl_crackle": |
| crackles = (rng.random(audio.shape[0]) > 0.996).astype(np.float32) * rng.uniform(-0.18, 0.18, audio.shape[0]) |
| audio = _lowpass(audio, sr, 12500) + crackles[:, None] * 0.25 |
| elif texture_id == "cassette_warmth": |
| audio = _compress(audio, 1.25) + rng.normal(0, 0.004, audio.shape).astype(np.float32) |
| audio = _lowpass(audio, sr, 11000) |
| elif texture_id == "fm_radio": |
| audio = _compress(_lowpass(audio, sr, 14000), 1.6) |
| elif texture_id == "mono_radio": |
| mono = audio.mean(axis=1, keepdims=True) |
| audio = np.repeat(mono, 2, axis=1) * 0.85 + audio * 0.15 |
| audio = _bandpass(audio, sr, 180, 5200) |
| elif texture_id == "club_pa": |
| audio = _compress(_echo(audio, sr), 1.55) |
| elif texture_id == "vhs_glow": |
| audio = _lowpass(_compress(audio, 1.2), sr, 9000) |
| elif texture_id == "mp3_compression": |
| audio = np.round(_lowpass(audio, sr, 8500) * 96) / 96 |
| return normalize_audio(audio) |
|
|
