| """Constant-gain volume normalization preserving shared timing.""" |
| import math |
| import numpy as np |
|
|
| def normalize_for_listening(wave, sample_rate, target_dbfs=-20., peak_dbfs=-1., max_boost_db=40.): |
| wave = np.asarray(wave, dtype=np.float32) |
| if wave.ndim != 1 or not np.isfinite(wave).all() or sample_rate <= 0: |
| raise ValueError('Expected finite mono audio and a positive sample rate') |
| if not all(math.isfinite(x) for x in (target_dbfs, peak_dbfs, max_boost_db)) or peak_dbfs >= 0 or max_boost_db < 0: |
| raise ValueError('Invalid normalization target/ceiling') |
| peak = float(np.max(np.abs(wave))) if len(wave) else 0. |
| if peak == 0: |
| return wave.copy(), dict(gain_db=0., input_peak=0., output_peak=0., active_rms=0., silent=True) |
| frame = max(1, round(sample_rate*.02)) |
| padded = np.pad(wave.astype(np.float64), (0, (-len(wave)) % frame)) |
| powers = np.square(padded.reshape(-1, frame)).mean(axis=1) |
| |
| active = powers >= max(float(powers.max())*10**(-35/10), 1e-20) |
| rms = float(np.sqrt(powers[active].mean())) if active.any() else 0. |
| if rms == 0: |
| return wave.copy(), dict(gain_db=0., input_peak=peak, output_peak=peak, active_rms=0., silent=True) |
| desired = 10**(target_dbfs/20)/rms |
| peak_gain = 10**(peak_dbfs/20)/peak |
| maximum = 10**(max_boost_db/20) |
| gain = min(desired, peak_gain, maximum) |
| normalized = wave*gain |
| return normalized, dict(gain_db=20*math.log10(gain), input_peak=peak, |
| output_peak=float(np.max(np.abs(normalized))), active_rms=rms, |
| output_active_rms=rms*gain, target_dbfs=target_dbfs, peak_ceiling_dbfs=peak_dbfs, |
| peak_limited=peak_gain < desired and peak_gain <= maximum, |
| boost_limited=maximum < desired and maximum < peak_gain, silent=False) |
|
|
