tape_recorder/core/compressor.py

import numpy as np

class Compressor:
    """
    A fast, block-based dynamic range compressor designed to run in real-time.
    Perfect for movie soundtracks to reduce dynamic range (makes quiet dialog louder, clamps loud explosions).
    """
    def __init__(self):
        self.enabled = False
        self.threshold_db = -20.0
        self.ratio = 4.0
        self.makeup_gain_db = 6.0
        
        # Smoothing factors for chunk-based envelope (0.0 to 1.0)
        self.attack_factor = 0.8   # Fast attack to catch peaks
        self.release_factor = 0.05 # Slower release for smooth pumping
        
        self._current_gain_db = 0.0
        
    def process(self, chunk: np.ndarray) -> np.ndarray:
        if not self.enabled or chunk.size == 0:
            return chunk
            
        # Calculate RMS of the entire chunk
        # Using peak detection or RMS. RMS is better for overall loudness, peak is better for hard limiting.
        # Movie dialogue is better managed with RMS.
        rms = np.sqrt(np.mean(chunk**2))
        rms_db = 20 * np.log10(max(rms, 1e-6))
        
        # Calculate target gain reduction
        if rms_db > self.threshold_db:
            overshoot = rms_db - self.threshold_db
            target_gain_db = -overshoot * (1.0 - 1.0 / self.ratio)
        else:
            target_gain_db = 0.0
            
        # Smooth the gain reduction to prevent clicks
        if target_gain_db < self._current_gain_db:
            # Gain is going down (compressing harder) -> Attack
            self._current_gain_db += (target_gain_db - self._current_gain_db) * self.attack_factor
        else:
            # Gain is going up (recovering) -> Release
            self._current_gain_db += (target_gain_db - self._current_gain_db) * self.release_factor
            
        # Combine with makeup gain and convert to linear scale
        total_gain_db = self._current_gain_db + self.makeup_gain_db
        linear_gain = 10 ** (total_gain_db / 20.0)
        
        # Multiply chunk (numpy vectorized)
        return chunk * linear_gain