traditional-speech-enhancement-algorithms

README.md的中文版见README_CN.md

This repository contains traditional speech enhancement algorithms, accompanying the book Speech enhancement_theory and practice.pdf. The PDF file is included in this repository and located in the same root directory as this README file. It includes classic speech enhancement algorithms and objective evaluation metrics for speech quality. This repository only includes matlab code and sample audio files (*.wav), corresponding to the MATLAB code section of the CD-ROM that comes with the book. The code was published quite some time ago and may require appropriate modifications to run on the latest versions of matlab. For example, in statistical_based/wiener_as.m:

wavwrite( enhanced_speech, fs, nbits, outfile);

needs to be changed to the following to run in matlab2016:

audiowrite(outfile,enhanced_speech,fs);

The overall folder structure is as follows:

+---spectral_subtractive                    # Spectral Subtraction Methods
+---statistical_based                       # Statistical-Based Methods
+---subspace                                # Subspace Methods
+---noise_estimation                        # Noise Estimation Algorithms
\---objective_measures                      # Objective Evaluation Metrics
|   +---intelligibility                     # Speech Intelligibility
|   \---quality                             # Speech Quality
+---ideal_channel_selection                 # Ideal Channel Selection

Each subfolder contains a readme.txt and readme.pdf, which provide usage examples and corresponding references for each algorithm. The algorithms included in each folder are as follows:

├─spectral_subtractive                      # Spectral Subtraction Methods
│      specsub.m                            # Basic spectral subtraction algorithm
│      mband.m                              # Multiband spectral subtraction algorithm
│      ss_rdc.m                             # Spectral subtraction with reduced delay convolution and adaptive gain averaging
│
├─statistical_based                         # Statistical-Based Methods
│      wiener_iter.m                        # Iterative Wiener filtering based on an all-pole speech model
│      wiener_as.m                          # Wiener filtering based on a priori SNR estimation
│      wiener_wt.m                          # Wiener filtering based on wavelet thresholding of multitaper spectra
│      mt_mask.m                            # Psychoacoustically-based algorithm
│      audnoise.m                           # Audible noise suppression algorithm
│      mmse.m                               # MMSE algorithm with or without speech presence uncertainty
│      logmmse.m                            # Log-MMSE algorithm
│      logmmse_SPU.m                        # Log-MMSE algorithm with speech presence uncertainty
│      stsa_weuclid.m                       # Bayesian estimator based on a weighted Euclidean distortion measure
│      stsa_wcosh.m                         # Bayesian estimator based on a weighted Cosh distortion measure
│      stsa_wlr.m                           # Bayesian estimator based on a weighted likelihood ratio distortion measure
│      stsa_mis.m                           # Bayesian estimator based on a modified Itakura-Saito distortion measure
│
├─subspace                                  # Subspace Methods
│      klt.m                                # Subspace method with embedded pre-whitening
│      pklt.m                               # Perceptually-based subspace method
│
├─noise_estimation                          # Noise Estimation
│      specsub_ns.m                         # A collection of different noise estimation algorithms
│      martin_estimation.m                  # Martin's minimum tracking algorithm
│      mcra_estimation.m                    # MCRA algorithm
│      mcra2_estimation.m                   # MCRA2 algorithm
│      imcra_estimation.m                   # Improved MCRA algorithm
│      doblinger_estimation.m               # Continuous minimum tracking algorithm
│      hirsch_estimation.m                  # Weighted spectral averaging
│      connfreq_estimation.m                # Joint time-frequency region
│
├─objective_measures                        # Objective Evaluation Metrics
│  ├─intelligibility                        # Speech Intelligibility
│  │      CSII.m                            # Computes the CSII measure
│  │      NCM.m                             # Computes the NCM measure
│  │      SII.m                             # Computes the SII measure
│  │
│  └─quality                                # Speech Quality
│          comp_snr.m                       # Computes segmental and overall SNR
│          comp_wss.m                       # Computes the weighted spectral slope measure
│          comp_llr.m                       # Computes the likelihood ratio
│          comp_is.m                        # Computes the Itakura-Saito measure
│          comp_cep.m                       # Computes the cepstral distance
│          comp_fwseg.m                     # Computes the frequency-weighted segmental SNR (fwSNRseg)
│          comp_fwseg_variant.m             # Computes a frequency-variant fwSNRseg
│          comp_fwseg_mars.m                # Computes a frequency-variant fwSNRseg based on MARS analysis
│          pesq.m                           # Computes PESQ (Perceptual Evaluation of Speech Quality)
│          composite.m                      # Computes the composite measures
│          addnoise_asl.m                   # Adds noise to clean speech at a specified SNR based on active speech level
│
└─ideal_channel_selection                   # Ideal Channel Selection
        ics.m                               # Channel selection based on SNR (ideal binary mask)
        ics_constr_rule.m                   # Channel selection based on the SNR_ESI rule
        ics_masker_rule.m                   # Channel selection based on masker constraints
        ics_reverb.m                        # Channel selection based on the SRR rule