Centeralized Speech Enhancement for Audio Conferencing

A preprocessing noise suppression algorithm using spectral subtraction has been developed, implemented, and tested.Spectral estimates for the background noise were obtained from the input signal during nonspeech activity. The algorithm can be implemented using a dual microphone source and requires a

Project Title

Centeralized Speech Enhancement for Audio Conferencing

Project Area of Specialization

Internet of Things

Project Summary

A preprocessing noise suppression algorithm using spectral subtraction has been developed, implemented, and tested.Spectral estimates for the background noise were obtained from the input signal during nonspeech activity. The algorithm can be implemented using a dual microphone source and requires about the same computation as a high-speech convolution. Its performance was demonstrated using short-time spectra with and without noise suppression and quantitatively tested for improvements in intelligibility and quality.

Project Objectives

Noise spectrum does create the annouying effect which is called Musical Noise. The spectral subtraction technique seems to allow us to remove a fair amount of Noise. Spectral subtraction seems to let us achieve better acoustic noise reduction.To achieve the Good estimate of Noise we take a signal that has some portion of only noise without any prior information to get a good estimate to design a filter.The other approach we can take is Voice active detection which can help us to find the Speech in a signal without explicitly identifying the portion which has only Noise because in a realistic situation Noise has a varying behavior.We want to design a simple algorithm  which descritizes the amplitude range of a noise so we can detect the most frequently occurring amplitude level to consider it as estimate of noise. We report our project on Centralized Speech enhancement for audio Conferencing. Our project targets the problem of Speech enhancement and all our experiment is on speech signal.

Project Implementation Method

• Installation of the MATLAB software.
• The parameters we used for audio signal recording

• Save the recorded signal and play the recorded file to check if the signal is correctly recorded or not.
• Human input: record the signal in such a way that

1. For first 3 seconds, noise signal is recorded.
2. For next 7seconds, the speech signal is recorded.

• System recordation: record the signal in the real-time.
• Recorded signal must have some noise which need to be removed at the receiver side. This noise may be added because of channel, environmental effects, other signals.
• In frequency domain it is easy to analyze the signal For the removal of noise,  we first convert the signal in the frequency domain using fast Fourier transform (FFT).
• FFT results in frequency spectrum of the signal.
• After this, using noise estimator we could estimate the added noise.
• Main objective is to remove that noise from the signal so that we have the original clear not attenuated audio recorded signal.
• For removing noise, many techniques are available..We have used frame by frame analysis on basis of Signal to Noise ratio (SNR)
• This analysis may have these 3 possibilities:

1. High SNR

    A ratio higher than 1 indicates more signal than noise.(speech)

     Step to take : Pass the signal.

1. Low SNR

    A ratio less than 1 indicates more noise than speech.(noise)

        Step to take : Remove noise from the signal.

1. Zero SNR

        Signal is pure.

• Time Domain (TD) conversion using inverse FFT, which results in audio signal in time domain.
• For verification result obtain from time domain conversion and the filter in time domain are equal.

By using the built-in MATLAB command “filter”.

• Future direction:
• To implement this design using Alsa-API in C-language.

Parameters

Sampling frequency

No of Channel

No of bits

Time duration

Benefits of the Project

Noise suppression is the method of eliminating the noise from a desired signal in order to enhance the audio quality in audio conferencing. The main objective of noise suppression is to reduce the noise during audio conferencing. By using Noise Suppression method it effects  the transmitted sound quality. Our algorithm is able to modify the audio stream but minor modification is not noticeable by human ear. For example

Lets clarify what noise suppression is. Noise suppression means suppressing the noise that goes from your background to the person you are having a call with and the noise coming from their background to you as figure shows

The background noise which comes from both sides. Noise suppression filters it out for both sides.This contrast with noise suppression which refers to suppress the noise which comes to your ears from surroundings.

Noise suppression has been effectively implemented in laptops and conferencing systems. The device captures the voice, once it captured. The designed algorithm filters it and the result sends to the receiver side.

Audio conferencing was quiet bad 10-20 years ago. Many algorithm was designed to improve the sound Quality.

Existing noise suppression solutions are not perfect to completely clean the voice from noisy atmosphere.

By the use of noise suppression algorithm it allows the receiver to hear the voice clearly. It benefits the user of audio conferencing and improves their ability to hear more clearly by overcome the noise and distraction.

Before the speech is transmitted to the receiver the designed system differentiate the noise and the speech then it suppressed the noise and transmitted the filtered noise at the receiver side. If you work in an open environment there is a much noise for the disturbance in a speech. It is better to use the designed algorithm for a better speech quality.

Our aim is to provide a high level of audio enhancement comfort during audio conferencing using noise suppression technique.

Technical Details of Final Deliverable

There is a one technique which we used to design the algorithm which will help us in this process. We record for 10 seconds in which some portion have  only noise and remaining have speech.

Noise is simply a room tone of a few seconds we can say the natural noise of the environment which we are recording (fan sound, heavy breathing, Birds sounds, system sound etc). Even if you can’t hear anything, a sensitive microphone will pick up ventilation noise, computer fans and more.

By taking room tone it will serve as a baseline for the software to suppress the noise. Having a portion of only noise in a recording is always a good practice.

We can also find only noise at the beginning of recording or end of a file. Where nothing much is happening usually that’s enough for the noise suppression designing.

Noise suppression based on some factors which are important to design the noise suppression algorithm.

Noise Suppression: Controls the amount of suppression of your noise volume.

Sensitivity: Controls the range of what noise removal considers noise. The higher this goes the more your actual audio (such as voices) will be affected.

ALTERNATIVES TECHNIQUES:

• Zero Crosssing

By using zero crossing technique we can identify the difference between noise and speech. We can differentiate them by the method of analyzing where zero crossing is fast means there is a noise and where zero crossing is slow means there is a speech.

• Voice Active Detection

Voice active detection is a designed algorithm which is used to detect the speech in a given signal. It differentiate the noise and speech in digital form. By applying the VAD algorithm on the recorded signal. It generates the binary file. 0 indicates noise and 1 indicates the speech.

Final Deliverable of the Project

Software System

Type of Industry

Media , Telecommunication

Technologies

Internet of Things (IoT)

Sustainable Development Goals

Industry, Innovation and Infrastructure

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com