Implementation and Performance Analysis of G.723.1 Speech Codec

Telephone networks normally transmit narrowband (NB) speech with a bandwidth restricted to 300 Hz to 3.4 kHz. While human speech comprises frequencies even further than the NB, there is a degradation of speech quality and intelligibility in such networks. This can be enhanced by recovering the speech components existing beyond the telephone bandwidth using Quadrature Mirror Filter (QMF) to give a perceptually superior wideband (WB) speech signal. QMF banks utilized as the fundamental mathematical tools to analyze and synthesize speech signals to coding of speech signal in sub-bands. Utilization of QMF's allows evading the aliasing effects due to samples decimation when signal is divide into sub-bands. In this paper, the performance of bandwidth extension (BWE) of speech through QMF was assessed by objective parameters like Log Spectral Distortion (LSD) and Itakura-Saito distance (ISD) and subjective parameters like Mean opinion score (MOS), Absolute Category Rating (ACR), Comparison Category Rating (CCR) and Perceptual Evaluation of Speech Quality (PESQ) which demonstrates that the recovered WB speech signal provides speech quality compared by original WB speech signal.

In today's wireless communication system, quality of decoded speech at receiving end is found muffled and thin, mainly attributed to inherent band limitation (300-3400 Hz) and power constraints. In order to obtain toll quality of recovered speech in terms of intelligibility and naturalness in wireless systems, Narrowband (NB) speech coders should be upgraded to its counterpart Wideband (WB) coders (50-7000Hz). In the meantime, a novel and backward compatible solution is proposed that claims to artificially extend bandwidth of NB speech to WB at receiving end, popular as Artificial Bandwidth Extension (ABE). Out of many techniques which aim to mitigate the effect of the ever unpredictable channel conditions, Adaptive Multi Rate (AMR) NB coder is considered to be one of the potential candidates. Selection of particular bit rate mode (out of all eight bit rate modes between 4.75 kbps and 12.2 kbps) solely depends upon the channel condition. This paper discusses development of ABE algorithm for Code Excited Linear Prediction (CELP) based GSM AMR NB coder, and for the same, MATLAB based e-test bench is created for simulation. Such series of simulations are conducted to discover and gain insight about the overall performance of proposed ABE coder that includes subjective (Mean Opinion Score - MOS) and objective (Perceptual Evaluation of Speech Quality - PESQ) analyses. The evaluated results for both analyses clearly advocate that proposed ABE coder outperforms legacy GSM AMR 06.90 NB coder.

Packet loss may affect a wide range of applications that use voice over IP (VoIP), e.g. video conferencing. In this paper, we investigate a time-domain convolutional recurrent network (CRN) for online packet loss concealment. The CRN comprises a convolutional encoder-decoder structure and long short-term memory (LSTM) layers, which have been shown to be suitable for real-time speech enhancement applications. Moreover, we propose lookahead and masked training to further improve the performance of the CRN framework. Experimental results show that the proposed system outperforms a baseline system using only LSTM layers in terms of two objective metrics – perceptual evaluation of speech quality (PESQ) and short-term objective intelligibility (STOI); it also reduces the word error rate (WER) more than the baseline when used as a frontend for speech recognition. The advantage of the proposed system is also verified in a subjective evaluation by the mean opinion score (MOS).

A mathematical matrix formalism which combines network management system (NMS) statistics from Cosmote GSM land mobile radio network and MOS measurements derived from drive tests are used, in order to extract a derivative MOS key performance indicator. Initially, the calculation is performed in a base station controller (BSC) area and then base transceiver stations (BTS) MOS distribution patterns are examined for different speech encodings. Examples of speech encoding effectiveness evaluation are performed using the previous speech encoding distribution patterns. Moreover, the effect of mobile station (MS) technology on different speech encodings is pointed out. Ultimately, different speech encodings are compared using quality and speech codec distribution patterns.

This paper presents the performance evaluation of 3G mobile networks for one kind of multimedia application called Voice over IP (VoIP) within Bangkok, the capital city of Thailand, using Line and Skype, two popular VoIP applications. This study used evaluation of voice quality provided by both applications. The tests have been conducted using stationary scenarios over 5 major 3G mobile networks, served by 5 operators, within 14 universities in the inner city of Bangkok in order to gather data of degraded speech files. Then, the data was measured using Perceptual Evaluation of Speech Quality (PESQ) to find Mean Opinion Score (MOS) before analyzing with ANOVA and T-test, which are statistical tools, so that the discussion and conclusion can be eventually derived.

The wide use of digital speech recorders becomes a serious matter when they are involved in assisting with court rulings. How to distinguish if a recorded content is valid or not becomes a life-or-death question. In light of this concern, least significant bits (LSB) of excitation signals would be used as fragile watermarks in the hybrid speech vocoder. In addition, a location-variable content-dependent watermark generating mechanism is proposed. Such location-variable content-based watermark would allow users to detect where in the recording the content is being replaced, inserted, or deleted. Lastly, an attempt is done to store partial reconstruction data in the LSBs of excitation signals in the G.723.1 speech codec, so that the original speech content may be reconstructed after counterfeited. The proposed system is demonstrated to be a reliable system, with test results showing that a recording with watermarks has a perceptual evaluation of speech quality (PESQ) value down 0.2, while the accuracy in detecting faked regions can be up to 97.45%.

Voice over Internet Protocol (VoIP) is a technology that transports voice data packets across packet switched networks using the Internet Protocol (IP). Perceived speech quality is the key metric for QoS for VoIP applications. The most popular objective measurement method for VoIP voice is E-Model. The primary aim of this work is to investigate the how the delay affect the VoIP quality and how to combined the affection in to standard E-Model. In this paper, a testing environment to measure the impairment of different delay jitter is setup. The Mean Opinion Scores (MOS) were calculated, Perceptual evaluation of speech quality (PESQ) algorithm is used. We use stand deviation of delay to describe the delay effect on voice quality, and give a new Ij formula, which is added to the standard E-model, as jitter impairment factor. Good accuracy is achieved by our new extended E-model.

Speech communicated is adversely affected by environmental noise. It is important to process the speech and reduce noise for better understanding. Speech enhancement or noise reduction is useful to provide comfort for human or machine listening. Traditional algorithms provide better noise reduction and better-quality speech. Due to the non-stationary nature of noise and the quasi-stationary nature of speech, the traditional methods are proven inadequate in achieving high-quality speech. Later statistical estimators based on Gaussian, and super-Gaussian Probability Density Function (PDF) assumption further improved the enhancement performance. But still, non-stationary noise nature introduces artifacts in processed signal and results in decreased performance. It is observed that neural network approaches and the factorization approach provide better performance even under non-stationary noises by proper training and large database. Different features result in variations in output performance under unseen noise and speaker conditions. It is important to understand the importance and advantages of traditional methods, statistical estimators, and neural network approaches performances. To select the suitable method for a required application, it is essential to consider the trade-off between quality and distortion. In this work, the importance of speech enhancement methods is discussed. Performance measures used for understanding the speech enhancement like Signal to Noise Ratio (SNR), Segmental SNR, Log-Likelihood Ratio (LLR), Weighted Spectral Slope (WSS), Perceptual Evaluation of Speech Quality (PESQ), Short-Time Objective Intelligibility (STOI), Signal to Distortion Ratio (SDR) and Mean Opinion Score (MOS), are given. Highlights of important results are discussed for analyzing better speech enhancement methods for the required application. In this work, performance is compared using objective and subjective performance measures. Simulation results show superior performance when neural network is employed in statistical estimators.

This paper investigates how the directional characteristics of an Acoustic Vector Sensor (AVS) can be used to separate speech sources. The technique described in this work takes advantage of the frequency domain direction of arrival estimates to identify the location, relative to the AVS array, of each individual speaker in a group of speakers and separate them accordingly into individual speech signals. Results presented in this work show that the technique can be used for real-time separation of speech sources using a single 20ms frame of speech, furthermore the results presented show that there is an average improvement in the Signal to Interference Ratio (SIR) for the proposed algorithm over the unprocessed recording of 15.1 dB and an average improvement of 5.4 dB in terms of Signal to Distortion Ratio (SDR) over the unprocessed recordings. In addition to the SIR and SDR results, Perceptual Evaluation of Speech Quality (PESQ) and listening tests both show an improvement in perceptual quality of 1 Mean Opinion Score (MOS) over unprocessed recordings.

Recent work in the domain of speech enhancement has explored the use of self-supervised speech representations to aid in the training of neural speech enhancement models. However, much of this work focuses on using the deepest or final outputs of self supervised speech representation models, rather than the earlier feature encodings. The use of self supervised representations in such a way is often not fully motivated. In this work it is shown that the distance between the feature encodings of clean and noisy speech correlate strongly with psychoacoustically motivated measures of speech quality and intelligibility, as well as with human Mean Opinion Score (MOS) ratings. Experiments using this distance as a loss function are performed and improved performance over the use of STFT spectrogram distance based loss as well as other common loss functions from speech enhancement literature is demonstrated using objective measures such as perceptual evaluation of speech quality (PESQ) and short-time objective intelligibility (STOI).

In this paper, localization and separation of acoustic sources are examined. Depending on the number of sources in relation to the array channels, the problem is investigated in terms of underdetermined and overdetermined configurations. In the underdetermined configuration, virtual monopole sources are assumed in uniformly spaced angles. The problem is then formulated into compressive sampling (CS) problem which can be solved by using the linearly constrained -norm convex (CVX) optimization. The solution yields the directions of real sources and the source signal spectrum, which enables localization and reconstruction of sources at one shot. In the underdetermined configuration, source localization and signal separation is carried out in two steps. First, the directions of arrival (DOA) are estimated with Minimum Variance Distortionless Response (MVDR) or Multiple Signal Classification (MUSIC). Next, Tikhonov regularization (TIKR) is utilized to recover the source spectrum. In the localization problem for both configurations, Neyman-Pearson detector is employed to determine thresholds for source detection. Numerical and experimental results show that the proposed methods produce improved speech quality in terms of mean opinion score (MOS) in perceptual evaluation of speech quality (PESQ) test.

This paper presents the details of the SRIB-LEAP submission to the ConferencingSpeech challenge 2021. The challenge involved the task of multi-channel speech enhancement to improve the quality of far field speech from microphone arrays in a video conferencing room. We propose a two stage method involving a beamformer followed by single channel enhancement. For the beamformer, we incorporated self-attention mechanism as inter-channel processing layer in the filter-and-sum network (FaSNet), an end-to-end time-domain beamforming system. The single channel speech enhancement is done in log spectral domain using convolution neural network (CNN)-long short term memory (LSTM) based architecture. We achieved improvements in objective quality metrics - perceptual evaluation of speech quality (PESQ) of 0.5 on the noisy data. On subjective quality evaluation, the proposed approach improved the mean opinion score (MOS) by an absolute measure of 0.9 over the noisy audio.

This paper describes the GSM enhanced full rate (EFR) speech codec that has been standardised for the GSM mobile communication system. The GSM EFR codec has been jointly developed by Nokia and University of Sherbrooke. It provides speech quality at least equivalent to that of a wireline telephony reference (32 kbit/s ADPCM). The EFR codec uses 12.2 kbit/s for speech coding and 10.6 kbit/s for error protection. Speech coding is based on the ACELP algorithm (algebraic code excited linear prediction). The codec provides substantial quality improvement compared to the existing GSM full rate and half rate codecs. The old GSM codecs lack wireline quality even in error-free channel conditions, while the EFR codec provides wireline quality not only for error-free conditions but also for the most typical error conditions. With the EFR codec, wireline quality is also sustained in the presence of background noise and in tandem connections (mobile to mobile calls).

Modern mobile communications require optimum bandwidth utilization with minimum loss, delay and good quality of speech transmission. This triggered the usage of low bit rate voice codecs, among which CELP codecs inherit the merits of both waveform and source codecs, like toll quality, low bit rate etc. In this paper several hardware and software optimization techniques are presented for efficient implementation of ITU G.729 standard (CS-ACELP, conjugate structure algebraic code excited linear prediction) of 8 Kbit/s bit rate on a real time digital signal processor (DSP), with the aim of overcoming the limitation of computational burden and also scaling this application for enhanced speed to process more channels. These techniques are in general applicable to any speech codec and DSP processor platform.

Advancement of digital signal processing and networking has raised many security and copyright concerns, thus it is very important to protect the authentication of digital data. In this work, an audio watermarking algorithm has been proposed which can be efficiently used for tamper detection and is also robust against reasonable attacks. Also, the watermarks are inaudible. The proposed algorithm can easily detect tampering as the watermarks are embedded at each frame without causing any audio degradation. In the proposed technique, first the audio signal is compressed using Graph Based Transform (GBT), for which watermarks are embedded into Line Spectral coefficients (LSFs) that are derived from linear prediction (LP) analysis with dither modulation-quantization index modulation (DM-QIM). Watermarks thus embedded in all frames are not only inaudible to the Human auditory system but also potentially provide robustness against meaningful attacks. This work also focuses on Blind tamper detection which is made effortless due to the proposed embedding algorithm. To measure the robustness of the algorithm, general processing of watermarked signals was done along with fragility testing. Quality of the audio was measured using Perceptual Evaluation of Speech Quality (PESQ) and Short-time objective intelligibility (STOI). The maximum PESQ score and STOI score of 2.8781 and 0.8150 respectively was observed without any attack on the audio signal. Tamper detection and quality measurement are the major contributions of this work. Detailed metric evaluation for attacks such as Scaling, Resampling, Filtering, Compression and Addition of White Gaussian noise (AWGN) has been computed and compared. The proposed technique makes tamper identification easier and gives framewise security.