An intelligent room acoustic correction system that automatically analyzes frequency response and generates optimal equalizer parameters for audio systems.
This project implements an advanced automatic EQ correction algorithm designed to optimize room acoustics for audio systems. Unlike traditional flat response approaches, our algorithm uses a tilted target curve that considers psychoacoustic principles and real-world listening preferences.
Example of automatic EQ correction showing original response (blue), target curve (red dashed), and corrected response (green)
- Intelligent Target Curve: Tilted frequency response (e.g., 3dB slope from 20Hz to 1kHz) for more natural sound
- Advanced Preprocessing: Two methods for handling problematic frequency response data
- Configurable Smoothing: Multiple octave-based smoothing options (1/3, 1/6, 1/12)
- Global Optimization: Uses simulated annealing for robust parameter optimization
- Perceptual Weighting: Logarithmic frequency weighting based on human auditory perception
- Automatic Parameter Generation: Outputs ready-to-use EQ filter parameters
pip install numpy scipy matplotlib- Configure your settings in the configuration section:
# Basic Configuration
DATA_FILE_PATH = 'your_frequency_response.txt'
TARGET_RANGE = [1000, 8000] # Reference frequency range (Hz)
CORRECTION_RANGE = [20, 1000] # Correction frequency range (Hz)
NUM_EQ_BANDS = 10 # Number of EQ filters
TARGET_SLOPE_DB = 3.0 # Target curve slope (dB)
# Processing Options
SMOOTHING_TYPE = '1/6' # Options: None, '1/3', '1/6', '1/12'
PREPROCESSING_METHOD = 1 # 1: Dip flattening, 2: Logarithmic correction- Run the algorithm:
python run_eq_analysis52.py- Get your results:
- EQ parameters printed to console
- Frequency response plot saved as PNG
- Ready-to-use filter settings
# Generate tilted target curve
log_f = np.log10(correction_freqs)
target_curve = np.interp(log_f, [log_f_20, log_f_1k], [level_at_20hz, level_at_1k])Method 1: Dip Flattening
dip_threshold_curve = target_curve - DIP_FLATTEN_THRESHOLD_DB
deep_dip_mask = analysis_mags < dip_threshold_curve
analysis_mags[deep_dip_mask] = target_curve[deep_dip_mask]Method 2: Logarithmic Correction
negative_mask = analysis_mags - target_curve < 0
analysis_mags[negative_mask] = target_curve[negative_mask] - np.log(1 + target_curve[negative_mask] - analysis_mags[negative_mask])def smooth_curve(freqs, mags, smoothing_type):
fractions = {'1/3': 3, '1/6': 6, '1/12': 12}
# Gaussian-weighted averaging within octave bands
weights = np.exp(-((np.log2(freqs[mask]/center_freq))**2))
smoothed_mags[i] = np.average(mags[mask], weights=weights)def cost_function(params, original_mags, target_curve, freq_axis, fs, log_freq_weights):
eq_curve_db = get_combined_eq_curve(params, freq_axis, fs)
corrected_mags = original_mags + eq_curve_db
squared_errors = (corrected_mags - target_curve)**2
weighted_squared_errors = squared_errors * log_freq_weights
return np.sqrt(np.mean(weighted_squared_errors))- Data Import: Load frequency response measurements
- Reference Calculation: Compute average level in target range (1-8kHz)
- Target Curve Generation: Create tilted reference curve
- Preprocessing: Apply selected correction method
- Smoothing: Apply octave-based smoothing for initial guess
- Optimization: Global optimization using preprocessed (non-smoothed) data
- Output: Generate EQ parameters and visualization
None: No smoothing applied'1/3': 1/3 octave smoothing'1/6': 1/6 octave smoothing'1/12': 1/12 octave smoothing
1: Dip Flattening - Clips deep notches to prevent over-correction2: Logarithmic Correction - Smooth handling of negative deviations
Files are automatically named based on configuration:
frequency_response_sloped_1-6_dip_flatten.pngfrequency_response_sloped_1-3_log_correction.png
The algorithm uses logarithmic frequency weighting to match human auditory perception:
log_freqs = np.log10(correction_freqs)
log_freq_widths = np.diff(mid_points, prepend=mid_points[0], append=mid_points[-1])- Frequency ordering: Center frequencies must be strictly increasing
- Gain limits: -30dB to +8dB range for practical applications
- Q factor range: 0.4 to 8.0 for usable filter characteristics
- Initial guess: Uses smoothed data for stability
- Final optimization: Uses preprocessed original data for detail preservation
Traditional flat response correction often sounds unnatural because:
- Psychoacoustics: Human hearing perceives slightly tilted curves as more balanced
- Room acoustics: Most rooms naturally boost low frequencies
- Musical preference: Slight low-frequency emphasis creates "warmth"
- Listening fatigue: Perfectly flat response can sound harsh over time
Your frequency response file should contain:
Frequency Magnitude
20.0 -2.5
25.0 -1.8
...
- Convergence: Typically converges within 1500 iterations
- Parameter quality: All generated parameters within practical ranges
- Listening experience: More natural sound compared to flat response
- Adaptability: Handles various room acoustic problems
# Modify target slope for different preferences
TARGET_SLOPE_DB = 2.0 # Gentler slope
TARGET_SLOPE_DB = 5.0 # More pronounced slope# Adjust based on room complexity
NUM_EQ_BANDS = 6 # Simple rooms
NUM_EQ_BANDS = 15 # Complex acoustic environmentsThis algorithm is based on:
- Psychoacoustic research on preferred listening curves
- Room acoustic theory for realistic correction limits
- Numerical optimization techniques for robust parameter finding
- Digital signal processing for accurate filter implementation
We welcome contributions! Areas for improvement:
- Adaptive filter count algorithms
- Full-range frequency correction
- Real-time processing capabilities
- Machine learning integration
This project is open source. Feel free to use, modify, and distribute according to your needs.
- Room measurement software
- Audio analysis tools
- Digital signal processing libraries
For questions, suggestions, or collaboration opportunities, please open an issue or reach out through the repository.
Keywords: Room Acoustics, EQ Correction, Audio Processing, Digital Signal Processing, Psychoacoustics, HiFi
Technologies: Python, NumPy, SciPy, Matplotlib, Signal Processing, Optimization Algorithms