auditory_models

Description

This repository provides multiple packages to compute auditory models, including

• Short Term Objective Intelligibility (STOI)
• Generalized Power Spectrum Model for audio quality (GPSMq)
• Binaural Auditory-Model-based Quality prediction (BAM-Q)

The repository is currently located here.

Prerequisites

Install Python >=3.9

Installation

pip install auditory_models

Usage

from auditory_models import GPSMq
import soundfile as sf

reference, fs_ref = sf.read("reference.wav")
degraded, fs_dgr = sf.read("degraded.wav")
if fs_ref != fs_dgr:
    raise ValueError("Sample rates must be equal!")

gpsmq = GPSMq()

gpsmq.process(reference, degraded, fs_ref)

This usage example can be applied to any other auditory model. Each model class is designed to only take keyword arguments so that they can be called in a default version without any arguments. The process() method is also standardized across all models to take reference, degraded, and sample_rate. Please refer to the models' individual documentation to learn more about their inputs and outputs.

Currently available models:

• BAMQ
• GPSMq
• STOI
• Base class for type hints: AuditoryModel

For each model-class there exists a frontend() and a backend() method. The process() method calls them both in succession, so frontend() and process() have the same input values and backend() and process() have the same output values. Since there is no universal definition of a front and a back end, we stick to the following idea here:

• frontend() is everything that extracts features from the audio data
• backend() is the statistical analysis of the features that results in the final value

Additional Info

• STOI differs from its original implementation in that the center-frequencies of the third-octave bands are fixed to the frequencies defined by IEC 61260-1:2014.
• There are some helpful implementations in ./auditory_models/helpers/ like a Gammatone filterbank, a Matlab-like resample class, etc.

Support

Regarding issues please feel free to contact me via zimmermannmax16@gmail.com

Contributing

Any contribution is welcome.

Authors and acknowledgment

Author: Max Zimmermann
Reviewer: Thomas Biberger
Credits to:

• The developers of the original Matlab implementations
  • STOI: Cees Taal
  • GPSMq and BAM-Q: Thomas Biberger and Jan-Hendrik Fleßner
• Manuel Pariente for the original Python implementation of STOI

License

This project is licensed under the GNU General Public License v3 (GPLv3). For further info see file COPYING.