Secure digital speech communication

• US 4,680,797 A
• Filed: 06/26/1984
• Issued: 07/14/1987
• Est. Priority Date: 06/26/1984
• Status: Expired due to Fees

• Current Assignee: The United States of America As Represented By The Secretary of Agriculture
• Original Assignee: United States Of America As Represented By The Secretary Of The Air Force
• Inventors: Benke, George
• Primary Examiner(s): Kemeny, E. S. Matt
• Assistant Examiner(s): Johnson, R. H.
• Application Number:
  US06/624,567
• Time in Patent Office: 1,113 Days
• Field of Search: 381/34-50, 381/29-50, 179/1.5 R, 179/1.5 G, 375/2.1, 375/122, 382/56, 455/26, 364/577, 364/723
• US Class Current: 704/211
• CPC Class Codes: G10L 25/00 Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 muting semiconductor-based amplifiers when some special characteristics of a signal are sensed by a speech detector, e.g. sensing when no signal is present, H03G3/34
• CPC Class Codes: H03M 1/1275 at extreme values only
• CPC Class Codes: H04B 14/046 Systems or methods for reducing noise or bandwidth

Abstract

A novel form of waveform coding is used, designated a "Critical Point Coder" (CPC). It operates by transmitting only those points in a waveform which are significant for defining its overall structure. The receiver reconstructs the missing points in the waveform using some sort of approximating interpolation. The Critical Point Coder uses a new approach to waveform coding. Rather than transmitting information about each sample as other waveform coders, the Critical Point Coder transmits information for only samples deemed critical to high quality reconstruction. The four major functions performed by the Critical Point Coder are (1) detecting which samples are critical, called critical points e.g. those points in the waveform where it is changing direction most rapidly, generally the peaks and valleys; (2) quantizing the vectors which are formed between successive critical points; (3) dealing with the non-uniform bit rate since critical points do not occur regularly; and (4) reconstructing the waveform between the endpoints of the quantized vectors. For stabilizing the bit rate, stabilization intervals of fixed duration are divided into equal analysis intervals, with logic to select from three modes for quantizing and encoding the vectors during each stabilization interval, depending upon the stepsize.

Claims

1. Signal processing apparatus using critical point coding comprising a transmitter and a receiver, which are interconnected via data transmission means for digitally encoded signals, the transmitter having an input from a voice signal source, and the receiver having a voice-signal output;

wherein the transmitter and receiver have corresponding vector tables, each defining precisely 32 regions which are necessary to quantize all of the vectors contained over a certain interval of time, designated an analysis interval which is equal to eight milliseconds, including bits for the table definition, with the table allowing for one to eight time samples and amplitude values of 0 to 256, to the range of the amplitude arising from two factors, (1) since the default for the vector slope sign is alternating, those vectors which have negative amplitude components are assigned the same regions as the positive amplitude vectors, and (2) with a compression ratio being used of 8;

1, the output of a 12-bit A/D converter is reduced from 2048 to 256;

whereby a typical 8-msec segment of voiced speech requires about 7 to 10 out of the 32 possible regions to quantize the vectors, and a typical segment of unvoiced speech requires only about 4 to 6 of these regions;

wherein the transmitter comprises;

(a) an analog filter having an upper cutoff frequency of about 3.6 kilohertz, said A/D converter, and sample means coupled in sequence from said voice signal source, to produce digital signals sampled at 16 kilobits per second using 12 bits per sample;

(b) compression means followed by adaptive filter means, with an input of the compression means coupled to said sample means, for compressing the data from the sample means sample-by-sample with a given compression function, for providing said compression ratio, and filtering adaptively according to the structure of the voice signal;

(c) buffer means coupled to the adaptive filter means for storing of digital samples of voice signals during each of successive stabilization intervals equal to 96 millseconds;

(d) analysis means for analyzing signals from the buffer means during each stabilization interval, with means for dividing each stabilization interval into twelve analysis intervals, and for each analysis interval finding those points in the waveform where it is changing direction most rapidly, these being the critical points, computing statistics, and defining a course table, the statistics being the number of critical points, the number of nonswitches (times when the vector slope does not change sign), and the number of regions of the vector table required for the vectors;

(e) logic means coupled to said analysis means for stabilizing the bit rate by calculating quantization stepsizes such that the number of bits required to code all the vectors is not exceeded, the logic means being effective during each stabilization interval to gather statistics about the signal structure, including the number of vectors, number of nonsign changes in vector slopes, and number of quantization regions, which are used to compute the quantization stepsize for the vectors, the stepsize being calculated in such a way that the bit rate for the entire stabilization interval is fixed at a given rate;

with there being three modes for handling the data, a first mode for when the stepsize is below a certain threshold value, with this stepsize being used to quantize all the vectors for the entire stabilization interval, a second mode for calculating the stepsize for the particular analysis interval, this stepsize being used for the vectors contained only for that analysis interval, and a third mode for use when the stepsize for the analysis interval is too large, coding being performed using an unvoiced mode;

(f) vector quantization means for quantizing the vectors using the stepsize calculated by said logic means (e);

(g) encoding means coupled to the vector quantization means for encoding the quantized vectors for each stabilization interval in a digit code using a format comprising a mode word defining which of said modes is being used, a stepsize word, and for each analysis interval, a table descriptor word followed by a sequence of the quantized vectors, including a special vector which indicates nonswitches and one vector which is a stop vector, with the first mode having only one mode word and stepsize word for the entire stabilization interval, and the second and third modes having a mode word and stepsize word for each analysis interval during the stabilization interval, and means for transmitting the resulting code via the data transmission means;

wherein the receiver comprises decode means coupled to receive signals from the data transmission means and to decode the bit stream, extraction means coupled to the decode means to extract the vectors, expansion means coupled to the extraction means to expand the vectors using the inverse of the compression function, and reconstruction means coupled between the expansion means and the voice-signal output using an interpolation function to construct the waveform between the critical points.

2. Signal processing according to claim 1, wherein the reconstruction means provides the properties that the interpolating function is monotonic between critical points, and for voiced speech the interpolating functions join smoothly between points.

3. Signal processing according to claim 2, wherein the number of entries in the vector tables is limited by replacing large vectors by a sum of at least two smaller vectors, limiting the time coordinates of vectors to a predetermined maximum and the amplitude to one-half of the total dynamic range.

4. Signal processing according to claim 1, wherein the reconstruction means provides the properties that the interpolating function is monotonic between critical points, and for voiced speech the interpolating functions join smoothly between points;

and

wherein the number of entries in the vector tables is limited by replacing large vectors by a sum of at least two smaller vectors, limiting the time coordinates of vectors to a maximum of two milliseconds and the amplitude to one-half of the total dynamic range.