Technical Field

The present invention relates to an encoder and decoder for analog signals, and to methods of encoding and decoding such signals. The invention has application, for example, to the field of digital transmission systems for transmitting speech signals, and to the field of conditioning speech signals so that they can be used as inputs to computerized systems.

Background Art

It is known to convert an analog speech signal into digital data, and vice versa, using clipped signal techniques.

A patent of interest for its teaching on clipped speech signal processing is U.S. Patent No. 4,477,925. In this patent, there is disclosed a system and a method which analyzes sampled clipped speech signals for purposes of identifying the original utterance. A sampler generates form the clippe signal a plurality of discrete binary values. A processor is used to analyze the sampled binary values to compare them against stored digitized signals corresponding to a known spoken utterance. Comparisons are made using linear predictive coefficients of an autocorrelation function of the utterances.

A problem that has been experienced with systems and methods employing clipped signal techniques is that clipped speech signal have a very flat sound with noise between words which is unnatural and fatiguing to listen to.

- 2 -

Disclosure of the Invention

It is an object of the present invention to provide apparatus and methods for alleviating the problem referred to above.

According to one aspect of the invention, there is provided an encoder including a first signal path for processing an analog signal, said first signal path including clipper means for clipping the analog signal to provide a clipped signal, and digitizer means for converting the clipped signal to a first digitized output signal, characterized by a second signal path for processing the analog signal, said second signal path including rectifying means for rectifying the analog signal, means for providing an amplitude signal which is a function of the amplitude of the rectified analog signal, and converter means for receiving the amplitude signal and for converting the amplitude signal to a second digitized output signal, said first and said second output signals together providing a representation of the analog signal.

The outputs of an encoder in accordance with this aspect of the invention, whether as two separate signal channels or as a single multiplexed channel, may be applied to a processing system for storage or analysis, or may be directed immediately to a decoder for reconstruction of the audio speech signal. (If the encoder outputs have been multiplexed into a single channel, then prior to decoding, the receiving end of the decoder will first provide the appropriate demultiplexing to recover the two digital channels.)

It should be appreciated that a speech encoder in accordance with the present invention operates to provide amplitude signal components in addition to a clipped digital signal, so as to enable a decoder to incorporate amplitude information in the reconstructed speech signal and thereby reconstruct

the naturalness and dynamic range of the original speech signal.

According to another aspect of the invention, there is provided a decoder characterized by digital- to-analog converter means having a first input for receiving a first digitized signal, which signal is a clipped signal representation of an analog signal, and a second input for receiving a second digitized signal, which signal represents amplitude variations of the analog signal, said converter means being arranged to convert the first digitized signal to an amplitude modulated signal as a function of the second digitized signal and to provide said amplitude modulated signal at an output, and means coupled to the output of said digital-to-analog converter means for shaping and filtering said amplitude modulated signal to provide .a reconstructed analog signal.

According to another aspect of the invention, there is provided a method of processing an analog signal including the steps of amplitude clipping the analog signal and digitizing the clipped analog signal to provide a first digitized output signal, characterized by the steps of rectifying the analog signal to provide a rectified analog signal, forming an envelope signal which is a function of the amplitude of the rectified analog signal, and digitizing the envelope signal to provide a second ditigized output signal which represents the amplitude components of the analog signal.

Brief Description of the Drawings

Embodiments of an encoder and decoder in accordance with the invention and of a system including such encoder and decoder will now be described by way of example with reference to the accompanying drawings, in which:

Fig. 1 is a block diagram of an encoder in accordance with the present invention;

Fig. 2 is a block diagram of a multiplexer/demultiplexer that may be used with a system for transmitting the encoded data in a single channel format convenient for computer input, output and storage;

Fig. 3 is a block diagram of a decoder in accordance with the present invention;

Figs. 4A - 41 illustrate a set of waveforms that represent signals taken at selected points (labeled SI through S9) in the encoder, and decoder block diagrams of Figs. 1 and 3;

Figs. 5A - 5B illustrate, in circuit diagram form, an implementation of the encoder of Fig. 1;

Fig. 6 illustrates, in circuit diagram form, the clock generator portion of the encoder implementation of Fig. 1;

Figs. 7A - 7D illustrate, in circuit diagram form, an implementation of the multiplexer/demulti¬ plexer of Fig. 2; and

Fig. 8 is a diagram of an implementation of the decoder of Fig. 3.

Modes for Carrying Out the Invention

Referring to Fig. 1, the analog speech signal (AUDIO INPUT shown as the waveform SI in Fig. 4A) is applied to the input of an audio amplifier/isolator 1. The output signal from the amplifier is directed to the input of an active low pass filter 10 which, in the preferred embodiment, cuts off signal frequency components above 3800 Hz and has a unity gain in the pass band. The low pass filter functions to remove unwanted components in the audio signal that lie outside of the effective frequency band of speech. It is also used to prevent or minimize the risk of incorrect analysis or recognition that could result

from the process of sampling. The filtered output signal is directed to a differentiator circuit 20 which is a high pass filter, with a 3 dB frequency at 16 kHz and an asymptotic slope of 6 dB per octave. The differentiator greatly improves the intelligibility of the clipped speech signal from what would otherwise be obtained from not differentiating. The differentiated signal S2, shown in Fig. 4B, from differentiator 20 is then applied to an infinite clipper 30. The infinite clipper 30 operates to produce a high or low level output signal S3, shown in Fig. 4C, depending on the polarity of the input at the time, and so it thereby preserves all the zero crossings of its input signal. The clipped output signal S3 is directed to a digitizer 40 which is clocked by a clocking signal CLKA, derived from a clock circuit 110, so as to uniformly sample the clipped output signal from the clipper 30. The output signal S4, shown in Fig. 4D, of digitizer 40 is thus a serial bit stream, which is a first digitized output signal of the encoder and which is directed to a first encoder output terminal labeled (£t In the preferred embodiment, the bit rate of the binary signal appearing at terminal (S) is 8000 bits per second, though rates above or below this value may be used; thus, the reconstructed speech from the decoder is quite intelligible down to a rate of 4000 bits per second. The encoder discussed so far constitutes a first signal path for the audio input signal. A second signal path is provided by the connection between the input to a full-wave rectifier 50 and the output from the differentiator 20. The rectifier performs a full-wave rectifying function upon its input signal and provides the rectified signal S5, shown in Fig. 4E, as an input to an envelope tracker 60. In the preferred embodiment of the invention, the envelope tracker has a fast attack (approximately 10

microsecond time constant) and slow decay (approximately 15 millisecond time constant). The output signal S6, shown in Fig. 4F, from the envelope tracker 60 is fed to an analog-to-digital converter 70 which is clocked by a clocking signal CLKB, of approximately 300 Hz or more, derived from the clocking circuit 110. The output signal S7, shown in Fig. 4G, generated by the converter 70 is a second digitized amplitude information signal which is directed to a second encoder output terminal, labeled ( ). The signals present at the output terminals ® and may now be processed in a number of manners, one of which is to store the signals in a computerized system for later recovery and decoding into an audio signal. Another possible application of the encoder output is for speech analysis or recognition. At this point it can be appreciated that the digitizing method for the amplitude information has been pulse code modulation (PCM). Other digitizing schemes may be utilized without detracting from applicants' contribution.

Referring to Fig. 2, the block diagram of an optional multiplexer/demultiplexer is shown, utilizing the clock signals CLKA, CLKB, and CLKC from the clock generator 110 of the encoder of Fig. 1. A power-up reset circuit 120 is provided to generate a reset signal RESET to the clock generator 110, the multiplexer 130 and the demultiplexer 140, in order to establish the system in a known state after power-up. The multiplexer 130 receives the two channel signals from the encoder terminals ©and |g)and multiplexes them into a single serial bit stream, labeled ©, which is synchronous to clock signal CLKC. The single channel signal may then be applied to a computer system 132 for storage or analysis via a serial input port. At some later time, the stored or analyzed signal may be retrieved from the computer system via a serial output port. The received serial signal Q, at

that time, is then directed to the demultiplexer 140 which recovers the two channel signals that were previously multiplexed and applies them to the output terminals (£)and(§

Referring now to Fig. 3, the block diagram of the decoder is illustrated, which decoder may be utilized in conjunction with the encoder of Fig. 1. The decoder receives the two signals from the encoder terminals (£)and (§), either directly from the encoder, or after storage, or after a multiplexing/de¬ multiplexing process. The input terminal receives the corresponding signal, labeled S4, originating from the encoder, which is the sampled clipped speech signal, and applies the signal to the reference terminal of a multiplying digital-to-analog converter 80. The input terminal ® receives the corresponding signal, labeled S7, originating from the encoder, which is the digitized speech amplitude signal (typically PCM), and applies the signal to the multiplying input of the multiplying digital-to-analog converter 80. The output of the converter 80 is thereby an amplitude modulated (or enhanced) sampled clipped speech signal S8, shown in Fig. 4H, referred to as AE-SCS. At this point, the AE-SCS signal corresponds to a differentiated speech signal because the encoder has performed a differentiation function prior to clipping and prior to envelope tracking and digitizing. The output signal from the converter 80 is directed to an integrator circuit 90 which removes any D.C. component from the signal and shapes the signal by an integration process which compensates for the differentiation process performed in he encoder. The output signal from the integrator 90 is then directed to a low pass filter 100 which removes signal frequency components above about 3800 Hz. The output of the low pass filter 100 is the reconstructed version of the audio input signal S9, shown in Fig.

- 8 -

41. That signal at this point may be directed to an audio amplifier and speaker for immediate review or it may be stored for future use or analysis.

Referring now to Figs. 5A and 5B, wherein a circuit diagram of the encoder is shown (minus the clocking circuit 110), an audio input speech signal, 2 volts peak-to-peak, is applied to the input of the audio amplifier 1. The amplifier 1 uses a Fairchild uA4136 operational amplifier and provides isolation and gain adjustment for the next stage, low pass filter 10, to which the output of the amplifier is directed. In the preferred embodiment, the low pass filter 10 consists of a two pole Butterworth active filter having a 3 dB cutoff frequency of about 3800 Hz, which was found to provide adequate out of band rejection for the environment in which the system was tested. The output of the filter 10. is directed to the differentiator circuit 20, which is a single pole high pass filter having a 3 dB corner frequency at 16 kHz and a gain of 10. The output of the differentiator 20 is directed to the input of the clipping circuit 30 and to the input of the full-wave rectifying circuit 50. These two separate signal paths will be described in turn.

The clipping circuit 30 in the preferred embodiment of the invention is comprised of an LM311 comparator connected as shown (Fig. 5B) . A If capacitor and a IK ohm resistor circuit on the front end of clipping circuit 30 are used to block D.C. A IK ohm pull-up resistor is used on the output end of the clipping circuit to adjust its output voltage to logic levels, approximately 0 to 5 volts, for compatibility with the following D-type flip-flop. The digitizer circuit 40 performs uniform sampling of the clipping circuit output, thereby providing the digitized output signal to the first encoder output terminal (§)L The signal will be referred to as the

- -

sampled clipped speech (SCS). The digitizer 40 consists of a CMOS 4013 D-type positive edge triggered flip-flop which samples the clipper output at a uniform rate determined by the clock signal CLKA provided by clock circuit 110, described later.

Referring back to the full-wave rectifier 50 (Fig. 5A) , the rectifier is comprised of three 4136 operational amplifiers, connected as shown, with diodes to perform a linear full-wave rectification on the signal received from the output of the differentiator 20. The output of the full-wave rectifier circuit 50 is directed to the envelope tracker circuit 60 which has as its central element a bilateral switch 4066 manufactured by National Semiconductor. If the input signal voltage of the tracker circuit exceeds the output of the tracker circuit (sensed by an operational amplifier 61), then the input (buffered by an operational amplifier 62) is switched into a charging circuit 63 on the output (Fig. 5B), causing the tracker circuit output to rise toward the input level with a time constant of about 10 microseconds. If the input voltage of the tracker circuit 60 falls below the output voltage, then the charging circuit 63 is switched out of the circuit path of operational amplifier 62, and the output is allowed to discharge and fall at a time constant of about 15 milliseconds. The envelope tracker circuit 60 thus provides a fast attack and slow decay envelope output which is directed to the input of the analog- to-digital converter circuit 70 which uniformly samples the envelope signal. In the preferred embodiment, the converter circuit 70 consists of an 8- bit successive approximation analog-to-digital converter device ADC 0804 manufactured by National Semiconductor. The output of the converter circuit 70 is presented on 8 bit lines which are coupled to the second encoder output terminal (§X The preferred

- 10 -

embodiment utilizes 8 bits without companding (or 7 bits if the multiplexer/demultiplexer of Figs. 2, 7A- 7D is used) , but fewer or more bits may be used and companding may be included without detracting from the present invention. The sampling rate of the converter circuit 70 is determined by the clock signal CLKB, of 30.0 Hz or more, provided by clock circuit 110.

Referring to Fig. 6, the clock circuit 110 consists of an oscillator 112 formed by two 74C14 Schmitt triggers and of two banks of synchronous down counters 114 and 116 formed with 74LS163 and 74LS08 chips which divide the oscillator frequency and produce the three clock signals CLKA, CLKB, and CLKC. The signal ENP is generated by a logic circuit network 148 of Figs. 7B and 7C.

Referring now to Figs. 7A-7D, wherein the circuit diagram of the multiplexer/demultiplexer 130/140 is shown, the power-up reset function is accomplished by the manually operated reset circuit 120. Depressing switch S 1 causes the RESET pulse to appear at the output of a 74LS08 gate. The multiplexer 130 receives the two output signals from the first and the second encoder terminals and <§}, respectively. The first serial bit stream signal is fed into serial- in, parallel-out, shift registers 131 and 132 as shown. The shift registers consist of SN74LS164 devices of the type manufactured by Texas Instruments. These registers are shifted by the rising edge of the clock signal CLKA. The parallel outputs of the registers 131 and 132 are connected to the parallel inputs of parallel-in, serial-out registers 134 and 135 as shown. The parallel loading of data is controlled by the output of a logic network 136 to avoid race conditions. The multiplexer 130 receives the signal on the input terminal <§)from the encoder into a parallel-in, serial-out register 133 as shown. Note that in this implementation, only 7 bit PCM is

accommodated; thus, the seven most significant bits from the encoder A/D converter 70 of Fig. 5B are used as inputs to the multiplexer register 133. The eighth bit of this register is tied high and is used as a check bit for synchronizing the demultiplexer, explained later. The parallel loading of data into register 133 is controlled by the logic network 136 mentioned previously. In the implementation shown, a parallel loading of data in registers 133, 134 and 135 occurs after receiving every 16 bits of the signal on the input terminal @. Af er each loading of the registers 133, 134 and 135, the data is shifted serially out by the clock CLKC, whose frequency is equal to the frequency of CLKA plus 8 times the frequency of CLKB. In this way, a steady serial bit stream, appearing at the terminal , is produced as the single channel output of the multiplexer. The output serial bit stream would typically be applied to a serial input port of a computer system for digital storage or analysis, as illustrated in Fig. 2. At some time later, the demultiplexer circuit 140 may retrieve the digital signal, at the input terminal (^ from the computer system via a serial port, which is clocked out by the applied clock signal CLKC from the demultiplexer. The retrieved bit stream is then fed to serial-in, parallel-out shift registers 141, 142 and 143 as shown. The registers are shifted by the rising edge of clock CLKC. The data of these registers is parallel loaded, as shown, into nontransparent latches 144 and 145 and shift registers 146 and 147. The loading is controlled by the logic network 148 which checks to see if there is a high bit value on pin 3 of register 141 just after the rising edge of clock CLKB. If the checked bit is low, then the data is not aligned according to the multiplexer format and so the logic network disables registers 144, 145, 146 and 147 until the checked bit becomes

high. In this way, the demultiplexer can eventually recover data synchronization if it is ever lost. If the checked bit is high, then the network issues a load signal to the registers 144, 145, 146 and 147, and enables their inputs. The seven parallel output bits of latches 144 and 145 correspond to the binary value of the signal applied to the terminal ®. The serial output of the shift register 147, clocked by CLKA, is the serial bit stream corresponding to the signal applied to the terminal .

Referring now to Fig. 8, wherein a circuit diagram of the decoder is shown, the two signals appearing at the terminals and ® are received by the multiplying digital-to-analog converter circuit 80 comprised of an AD7520 device manufactured by National Semiconductor and an operational amplifier 4136. The signal at the terminal <^is the sampled clipped speech signal and is used as the analog reference to pin ^' 15 of the AD7520 device, after having its D.C. component removed by an input RC filter formed by a .If capacitor and a 100K resistor. The 8 bit lines representing the binary value of the speech amplitude envelope are directed to pins 4 through 11 of the AD7520 device as shown. (Note, however, that if the multiplexer/demultiplexer of Fig. 6 is used, then the signal on the terminal §) consists of only 7 bit PCM and so pin 11 of AD7520 is tied low.) An operational amplifier 4136 is connected as shown to provide the output signal from the converter circuit 80, which is an amplitude enhanced form of the sampled clipped speech signal. The output signal from circuit 80 is directed to the integrator circuit 90 which is an active low pass filter that also blocks D.C. and suppresses frequencies below about 80 Hz. The active element of the circuit 90 is an operational amplifier 4136. The integration function is performed in the circuit by providing a -6 dB per octave slope across

the speech band frequencies above about 160 Hz. The output of integrator circuit 90 is directed to circuit 100 which is a single pole unity gain active low pass filter with a 3 dB corner frequency of 4000 Hz that utilizes an operational amplifier 4136. The output from the low pass filter 100 is the reconstructed audio system signal.