Variable Rate Speech Research Articles

Variable-rate CELP based on subband flatness

Code-excited linear prediction (CELP) is the predominant methodology for communications quality speech coding below 8 kbps, and several variable-rate CELP schemes have been discussed in the literature, including QCELP, the variable-rate wideband digital cellular mobile radio speech coding standard specified in IS-95. A key component of these speech coders is the detection and classification of speech activity, and several cues for rate variation have been studied, such as measuring the short-term speech energy, deciding whether the speech is voiced or unvoiced, or making more sophisticated phonetic classifications. We present a new method for rate variation based on a measure of subband spectral flatness, called spectral entropy. Spectral entropy is a normalized indicator of the texture of the input spectrum and is thus less dependent on speech and background noise energy variations. We present some results on the use of spectral entropy for voice activity detection across subbands and then evaluate using spectral entropy for deriving mode and rate allocation cues for a variable-rate CELP coder operating at an average rate of 2 kbps. To achieve communications quality speech at this rate, we develop a new split-band vector quantization (VQ) technique for representing the line spectral pairs and a multiple codebook approach for efficiently quantizing the coefficients of a three-tap pitch predictor, called lag-indexed VQ.

Neural dynamics of variable-rate speech categorization.

What is the neural representation of a speech code as it evolves in time? A neural model simulates data concerning segregation and integration of phonetic percepts. Hearing two phonetically related stops in a VC-CV pair (V = vowel; C = consonant) requires 150 ms more closure time than hearing two phonetically different stops in a VC1-C2V pair. Closure time also varies with long-term stimulus rate. The model simulates rate-dependent category boundaries that emerge from feedback interactions between a working memory for short-term storage of phonetic items and a list categorization network for grouping sequences of items. The conscious speech code is a resonant wave. It emerges after bottom-up signals from the working memory select list chunks which read out top-down expectations that amplify and focus attention on consistent working memory items. In VC1-C2V pairs, resonance is reset by mismatch of C2 with the C1 expectation. In VC-CV pairs, resonance prolongs a repeated C.

Neural dynamics of variable-rate speech categorization.

Neural dynamics of variable-rate speech categorization.

Capacity enhancement of cellular CDMA by traffic-based control of speech bit rate

Variable rate speech coding is now recognized as an important system component for high-capacity cellular networks because it exploits speech statistics to reduce the average bit rate, which results in reduced interference and increased capacity. Once a variable rate capability is available, an additional capacity enhancement can be achieved by introducing network control of the user bit rate in response to changing traffic levels. We introduce the concept of network control of rate and propose a particular network-control method for code-division multiple access (CDMA) systems. Based on an M/M//spl infin///M queueing model applied to a cell under heavy traffic conditions and a new performance measure called averaged speech quality, we obtain simulation results to demonstrate how network control of rate can achieve improved speech quality or increased capacity for a given quality objective.

A variable rate multipulse speech coder for CDMA cellular systems

Variable Rate (VR) speech coders are classified into: source-controlled VR coders where the rate is selected depending on the local character of the speech, and network-controlled VR coders where an external control signal selects the coding rate. The first category benefits from the variable rate channels used by Code Division Multiple Access (CDMA) mobile communications. The second category is indispensable for the right behaviour of the CDMA systems under conditions as high traffic levels. The VR speech coder presented in this communication exhibits both types of control. The source control is achieved by means of a Voice Activity Detector (VAD) and a phonetic classifier. The network control acts on the selection procedure of the multipulse excitation sequence to the synthesis filter. This is the main advantage of our VR MultiPulse speech coder because by means of an external signal the bit rate can be changed only every 4 msec, without transitions or distortions. Considering one-way communication, six different operating rates can be externally selected ranging from 4.8 to 9.1 kbps for the active frames; an average bit rate of 380 bps is required for the noise frames.

Real-time testbed for assessing a CDMA-based system

A CDMA-based system concept has been developed with a flexible air interface. A real-time testbed has been built based on this advanced, flexible concept demonstrating some of its key features. Variable bit rate speech up to 16 kb/s and data up to 128 kb/s in the connection-oriented and connection-less modes can be demonstrated. The testbed is used in a validation program assessing how well the CODIT (Code Division Testbed) concept meets the requirements set up for UMTS/FPLMTS. Laboratory tests verify basic performance of the CODIT air interface. This forms the basis for the overall performance of the CODIT system. Demonstrations verify performance in a set of real environments. The results of the validation process, including simulations, measurements, and theoretical studies form a basis for the final assessment of how well the CODIT concept is supporting UMTS. >

Method and system for transmitting variable rate speech signal

A speech signal transmission system for transmitting coded speech signals with variable bit rate is disclosed, which comprises a coder for analyzing digital speech signals inputted in a one-frame period and transforming them into coded data comprising a plurality of parameters indicating characteristics of the inputted speech signals, a data arranging circuit for arranging the coded data outputted by the coder in the order of the priority in the decoding of the speech signals and outputting them, and a bit stealer allowing a series of coded data outputted successively by the data arranging circuit to pass through only in a period of time determined by the transmission bit rate. The bit arranging circuit outputs the parameters in the order of the priority of decomposes each of them in unit of a bit and outputs them with decreasing priority. Parameters or bits of low priority are omitted by the bit stealer, depending on the transmission bit rate. One the receiver side the parameters are extracted and the speech signals are decoded on the basis of arrangement type identification codes transmitted together with the coded data.

Working Memory Networks for Learning Temporal Order with Application to Three-Dimensional Visual Object Recognition

Working memory neural networks, called Sustained Temporal Order REcurrent (STORE) models, encode the invariant temporal order of sequential events in short-term memory (STM). Inputs to the networks may be presented with widely differing growth rates, amplitudes, durations, and interstimulus intervals without altering the stored STM representation. The STORE temporal order code is designed to enable groupings of the stored events to be stably learned and remembered in real time, even as new events perturb the system. Such invariance and stability properties are needed in neural architectures which self-organize learned codes for variable-rate speech perception, sensorimotor planning, or three-dimensional (3-D) visual object recognition. Using such a working memory, a self-organizing architecture for invariant 3-D visual object recognition is described. The new model is based on the model of Seibert and Waxman (1990a), which builds a 3-D representation of an object from a temporally ordered sequence of its two-dimensional (2-D) aspect graphs. The new model, called an ARTSTORE model, consists of the following cascade of processing modules: Invariant Preprocessor → ART 2 → STORE Model → ART 2 → Outstar Network.

Finite state CELP for variable rate speech coding

The performance of a variable rate code excited linear predictor system is investigated. The coding system is based on a finite state CELP (FSCELP) frame work. Each individual state is primarily identified with a LPC model order, LPC coefficients bit allocation, excitation code book population density and state encoding rate. Successive input speech vectors are encoded at a rate that depends on the current state of the FSCELP system and the input vector characteristics. The use of a finite state system involves implicit clustering of speech signals. The lower rate states are selected during highly correlated steady state speech segments when relatively few bits are required to obtain adequate fidelity. For speech signals with a strong glottal excitation, unvoiced signals and transient speech segments, a relatively greater quantisation accuracy is needed to obtain good fidelity and therefore higher rate states of the system are used. Further improvement is obtained by using gamma populated excitation codebooks, for those states that are mainly used to encode speech signals with a strong underlying glottal excitation pulses. Experiments focus on investigation of the varying encoding requirements of the excitation signal for low pass, voiced, unvoiced and transient speech signals. The parameters of the finite state CELP system are designed to match the encoding requirements of typical speech signals. The greater part of the coding gain is obtained from variable rate encoding of the excitation signal. Using a six-state FSCELP, good quality speech is obtained at an average, maximum and minimum bit rates of 4 kbit/s, 10 kbit/s and 2 kbit/s, respectively.

Real-time speech segmentation using pitch and convexity jump models: application to variable rate speech coding

Convexity jump models are combined with period identification in the time domain to provide an efficient segmentation technique adapted to the needs of coders. The algorithm presented works online, with no training, and is speaker-independent. Structural problems found in previous systems (blank detection spots, long decision delays) have been overcome. Both the simplicity and the generality (information theory basis for the divergence test) allow applications of the automatic online segmentation method in other areas of signal processing. >

Variable rate speech coding for asynchronous transfer mode

A new speech coding and multiplexing scheme matched to the asynchronous transfer mode is described. A block coding technique that is based on a variable-rate coding algorithm that makes the most of the burstiness of voice information is employed. The main feature of the scheme is considerable bit reduction, which is attained by a fairly simple algorithm. It is demonstrated that the proposed algorithm exhibits better quality than that of a 32 kb/s ADPCM at a mean bit rate of less than 13 kb/s. The effect of statistical multiplexing is verified by means of simulation employing long conversational speech samples. Methods for constructing variable- and fixed-length frames (units of information multiplexed and transferred in the network) are proposed. The proposed coding algorithm is shown to be applicable to both variable- and fixed-length frame strategies.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Performance of packet reservation multiple access for digital mobile radio using variable rate speech coder

The advantages of using adaptive bit rate voice coders to improve the performance of PRMA for digital cellular radio are considered. It is shown that such a technique forces a fairer distribution of packet loss between mobiles and is able to accommodate fluctuations in traffic levels within a cell.

Variable rate speech compression by encoding subsets of the PARCOR coefficients

In LPC analysis, the speech signal is divided into frames each of which is represented by a vector of estimated vocal tract parameters, assumed to be constant throughout the frame. For many sounds, these parameters do not change significantly from one frame to the next, and some of them can often be adequately represented by previously transmitted values. In the LPC coding systems described in this paper, a number of alternative representations are considered for each frame. These representations (vectors) are combinations of PARCOR coefficients from the current frame and from previous frames. Several consecutive frames are analyzed at once, and all the possible sequences of PARCOR coefficient vectors are examined. The sequence which minimizes a preselected cost function is chosen for transmission, resulting in a reduced overall data rate. The examination of all the decision sequences is equivalent to a decision tree search, which is most efficiently accomplished through dynamic programming. Using these techniques, LPC encoded speech at 1200 bits/s is demonstrated to be of quality comparable to a constant rate LPC vocoder at 2400 bits/s.

Variable Rate ADPCM Based on Explicit Noise Coding

This paper discusses a variable bit rate speech coding system based on explicit coding of the reconstruction noise in ADPCM (differential pulse code modulation with adaptive quantization). If the ADPCM bit rate is R bits/sample, PCM coding of its noise using an average bit rate of R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> bits/sample provides the receiver with the possibility of operating at any bit rate in the range R to R + max{R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> }. Using R values in the range 2 to 5, and R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> values in the range 0 to 3, we compare the performance of the (R + R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> )-bit system with that of conventional (R + R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> )-bit ADPCM. If noise coding is based on instantaneous R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> -bit quantization of its samples with an optimized step size, the signal-to-noise ratio performance is comparable to that of conventional ADPCM for R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> = 1, but it deteriorates significantly for R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> > 1. With non-instantaneous noise coding, the performance can exceed that of conventional ADPCM for any R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> > 1, if R > 2. This is due to a variable bit allocation algorithm that quantizes noise samples with differing resolutions, while maintaining a constant total bit rate in every block of 4 ms. The algorithm does not require the transmission of any extra side information. It can also be regarded as a way of improving the performance of ADPCM coding at a single bit rate of R + R <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">n</inf> bits/sample.