Speech analysis and synthesis system

A speech analysis and synthesis system comprising:

• an analyzer including,
• means for analyzing an electrical speech signal by Cepstrum technique to generate a pitch period parameter,
• means for analyzing said speech signal to generate a voiced/unvoiced decision parameter, and
• means for analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• and
• a synthesizer including,
• pitch pulse generator means for receiving said pitch period parameter and generating pitch pulses having a corresponding pitch period,
• random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced, and from said random noise generator means if said decision is unvoiced,
• gain control means connected to said output of said switch means for receiving said power parameter, an output level of said gain control means being dependent upon said power parameter, and
• linear prediction filter means connected to said output of said gain control means for receiving and applying said coefficient parameters to said filter means to generate a replica or said speech signal at an output of said filter means;
• wherein said means for analyzing said speech signal to generate a voiced/unvoiced decision parameter includes
• means for scanning a Cepstrum signal within an auxilliary signal range on each side of a Cepstrum peak amplitude signal,
• means for detecting and storing a range peak signal within the scanned auxilliary ranges,
• means for summing the absolute values of the signal within the auxilliary ranges to form a range signal sum,
• means for dividing said Cepstrum peak amplitude signal by said range peak signal to form a range peak factor,
• means for dividing said range signal sum by said Cepstrum peak amplitude signal to form an average peak factor,
• means for first comparing said range peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said range peak factor is above said upper limit and generating an unvoiced decision parameter if said range peak factor is equal to or below said lower limit, and
• means for secondarily comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A speech analysis and synthesis system comrising:

• an analyzer including,
• means for analyzing an electrical speech signal by Cepstrum technique to generate a pitch period parameter,
• means for analyzing said speech signal to generate a voiced/unvoiced decision parameter, and
• means for analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• and
• a synthesizer including,
• pitch pulse generator means for receiving said pitch period parameter and generating pitch pulses having a corresponding pitch period,
• random noise generator means,
• switch means connected to said pitch pulsed generator means to said random noise generator means for receiving said voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced, and from said random noise generator means if said decision is unvoiced,
• gain control means connected to said output of said switch means for receiving said power parameter, an output level of said gain control means being dependent upon said power parameter, and
• linear prediction filter means connected to said output of said gain control means for receiving and applying said coefficient parameters to said filter means to generate a replica or said speech signal at an output of said filter means;
• wherein said means for analyzing an electrical speech signal by Cepstrum technique includes means for conditioning a Cepstrum signal generated by said Cepstrum technique, said signal conditioning means generating a weighting signal which linearly increases in value during the time span of said Cepstrum and said means for conditioning said Cepstrum further adding said weighting signal to said Cepstrum during said time span to create a weighted Cepstrum signal.

A method of analyzing and synthesizing, in real time, speech signals in a plurality of speech channels, said method comprising the steps of:

• analyzing said speech signals by,
• analyzing each of said plurality of speech signals by a Cepstrum technique to generate a pitch period parameter,
• analyzing each of said plurality of speech signals to generate a voiced/unvoiced decision parameter, and
• analyzing each of said pluality of signals by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• and
• synthesizing said speech signals by,
• generating pitch pulses having a pitch period corresponding to said pitch period parameter generated in said first analyzing step,
• generating a random noise signal,
• switching the output of said generated pitch pulses to an output of a switch means if said voiced/unvoiced decision parameter generated during said second analysis step is voiced and switching said random noise signal to said output of said switch means if said voiced/unvoiced decision is unvoiced,
• controlling the gain of said output of said switch means with gain control means in accordance with said generated power parameter, and
• connecting the output of said gain control means to the input of linear prediction filter means and applying said coefficient parameters to said filter means to generate a replica of each of said plurality of speech signals at an output of said filter means.

A method of analyzing and synethesizing speech signals comprising the steps of:

• analyzing said speech signals by, analyzing said electrical speech signal by a Cepstrum technique to generate a pitch period parameter,
• analyzing said speech signal to generate a voiced/unvoiced decision parameter,
• analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• and p1 synthesizing said speech signal by,
• generating pitch pulses having a pitch period corresponding to said pitch period parameter generated in said first analyzing step,
• generating a random noise signal,
• switching the output of said generated pitch pulses to an output of a switch means if said voiced/unvoiced decision parameter generated during said second analysis step is voiced and switching said random noise signal to said output of said switch means if said voiced/unvoiced decision is unvoiced,
• controlling the gain of said output of said switch means with gain control means in accordance with said generated power parameter, and
• connecting the output of said gain control means to the input of linear prediction filter means and applying said coefficient parameters to said filter means to generate a replica of said speech signal to an output of said filter means;
• wherein said step of analyzing said speech signal to generate a voiced/unvoiced decision parameter includes
• scanning a Cepstrum signal within an auxilliary signal range on each side of a Cepstrum peak amplitude signal,
• detecting and storing a range peak signal within the scanned auxilliary ranges,
• summing the absolute values of the signals within the auxilliary ranges to form a range signal sum,
• dividing said Cepstrum peak amplitude signal by said range peak signal to form a range peak factor;
• dividing said range signal sum by said Cepstrum peak amplitude signal to form an average peak factor,
• first comparing said range peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said range peak factor is above said upper limit and generating an unvoiced decision parameter if said range peak factor is equal to or below said lower limit, and
• secondarily comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A method of analyzing and synthesizing speech signals comprising the steps of:

• analyzing said speech signals by,
• analyzing said electrical speech signal by a Cepstrum technique to generate a pitch period parameter,
• analyzing said speech signal to generate a voiced/unvoiced decision parameter,
• analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• and
• synthesizing said speech signal by,
• generating pitch pulses having a pitch period corresponding to said pitch period parameter generated in said first analyzing step,
• generating a random noise signal,
• switching the output of said generated pitch pulses to an output of a switch means if said voiced/unvoiced decision parameter generated during said second analysis step is voiced and switching said random noise signal to said output of said switch means if said voiced/unvoiced decision is unvoiced,
• controlling the gain of said output of said switch means with gain control means in accordance with said generated power parameter, and
• connecting the output of said gain control means to the input of linear prediction filter means and applying said coefficient parameters to said filter means to generate a replica of said speech signal to an output of said filter means;
• wherein said step of analyzing electrical speech signals by Cepstrum technique includes a signal conditioning step for conditioning a Cepstrum signal which has a dimension in the direction of an ordinate axis, which is the amplitude of said Cepstrum signal and has a dimension in the direction of the abscissa which is a quantity representing time, said signal conditioning step having the steps of,
• generating a weighting signal which linearly increases in value with the dimension in the direction of the abscissa corresponding to said quantity representing time of said Cepstrum, and
• adding said weighting signal to said Cepstrum signal along said dimension in the direction of the abscissa thereof to generate a weighted Cepstrum signal.

A method of analyzing and synthesizing speech signals comprising the steps of:

• analyzing said speech signals by,
• analyzing said electrical speech signal by a Cepstrum technique to generate a pitch period parameter,
• analyzing said speech signal to generate a voiced/unvoiced decision parameter,
• analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• and
• synthesizing said speech signal by,
• generating pitch pulses having a pitch period corresponding to said pitch period parameter generated in said first analyzing step,
• generating a random noise signal,
• switching the output of said generated pitch pulses to an output of a switch means if said voiced/unvoiced decision parameter generated during said second analysis step is voiced and switching said random noise signal to said output of said switch means if said voiced/unvoiced decision is unvoiced,
• controlling the gain of said output of said switch means with gain control means in accordance with said generated power parameter, and `connecting the output of said gain control means to the input of linear prediction filter means and applying said coefficient parameters to said filter means to generate a replica of said speech signal to an output of said filter means;
• wherein said pitch pulse generating step includes
• generating a sequence of control signals in response to receipt of said pitch period parameter, and
• generating a substantially periodic, amplitude increasing pitch pulse excitation signal in response to said sequence of control signals.

A system for analyzing a plurality of speech signals in real time and generating representative control parameters therefor, said system comprising:

• means for analyzing a plurality of electrical speech signals by Cepstrum technique to generate a pitch period parameter for each of said signals;
• means for analyzing each of said plurality of speech signals to generate a voiced/unvoiced decision parameter;
• means for analyzing each of said plurality of speed signals by linear prediction technique to generate a predetermined number of coefficent parameters and a power parameter;
• and
• means for combining said pitch period parameter, said voiced/unvoiced decision parameter, said coefficient parameters and said power parameter into a composite signal representative of each of said plurality of speech signals for transmission in compressed form to a point of utilization.

A system for analyzing a speech signal and generating representative control parameters therefor, said system comprising:

• means for analyzing an electrical speech signal by Cepstrum technique to generate a pitch period parameter;
• means for analyzing said speech signal to generate a voiced/unvoiced decision parameter;
• means for analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameters;
• and
• means for combining said pitch period parameter, said voiced/unvoiced decision parameter, said coefficient parameters and said power parameter into a composite signal representative of said speech signal;
• wherein said means for analyzing said speech signal to generate a voiced/unvoiced decision parameter includes
• means for scanning a Cepstrum signal within an auxilliary signal range on each side of a Cepstrum peak amplitude signal,
• means for detecting and storing a range peak signal within the scanned auxilliary ranges;
• means for summing the absolute values of the signals within the auxilliary ranges to form a range signal sum,
• means for dividing said Cepstrum peak amplitude signal by said range peak signal to form a range peak factor,
• means for dividing said range signal sum by said Cepstrum peak amplitude signal to form an average peak factor,
• means for first comparing said range peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said range peak factor is above said upper limit and generating an unvoiced decision parameter if said range peak factor is equal to or below said lower limit, and
• means for secondarily comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A system for analyzing a speech signal and generating representative control parameters therefor, said system comprising:

• means for analyzing an electrical speech signal by Cepstrum technique to generate a pitch period parameter;
• means for analyzing said speech signal to generate a voiced/unvoiced decision parameter;
• means for analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• and
• means for combining said pitch period parameter, said voiced/unvoiced decision parameter, said coefficient parameters and said power parameter into a composite signal representative of said speech signal;
• wherein said means for analyzing an electrical speech signal by Cepstrum technique includes a signal conditioning system for conditioning a Cepstrum signal which has, in the direction of an ordinate axis, an amplitude dimension of said Cepstrum signal and, in the direction of the abscissa, a quantity representing time, said signal conditioning system having,
• means for generating a weighting signal which linearly increases in value with the dimension in the direction of the abscissa corresponding to said quantity representing time of said Cepstrum, and
• means for adding said weighting signal to said Cepstrum signal along said dimension in the direction of the abscissa thereof to generate a weighted Cepstrum signal.

A method for analyzing a plurality of speech signals in real time and generating representative control parameters therefor, said method comprising the steps of:

• analyzing each of a plurality of speech signals by a Cepstrum technique to generate a pitch period parameter;
• analyzing each of said plurality of speech signals to generate a voiced/unvoiced decision parameter;
• and
• analyzing each of said plurality of speech signals by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter.

A method for analyzing a speech signal and generating representative control parameters therefor, said method comprising the steps of:

• analyzing said speech signal by a Cepstrum technique to generate a pitch period parameter;
• analyzing said speech signal to generate a voiced/unvoiced decision parameter;
• and
• analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• wherein said step of analyzing said speech signal to generate a voiced/unvoiced decision parameter includes
• scanning a Cepstrum signal within an auxilliary signal range on each side of a Cepstrum peak amplitude signal,
• detecting and storing a range peak signal within the scanned auxilliary ranges,
• summing the absolute values of the signals within the auxilliary ranges to form a range signal sum,
• dividing said Cepstrum peak amplitude signal by said range peak signal to form a range peak factor,
• dividing said range signal sum by said Cepstrum peak amplitude signal to form an average peak factor,
• first comparing said range peak factor with predetermined upper and lower limits and generating an voiced decision parameter if said range peak factor is above said upper limit and generating an unvoiced decision parameter if said range peak factor is equal to or below said lower limit, and
• secondarily comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A method for analyzing a speech signal and generating representative control parameters therefor, said method comprising the steps of:

• analyzing said speech signal by a Cepstrum technique to generate a pitch period parameter;
• analyzing said speech signal to generate a voiced/unvoiced decision parameter;
• and
• analyzing said speech signal by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter;
• wherein said analyzing electrical speech signal by Cepstrum technique includes the step of conditioning a Cepstrum signal, which has a dimension in the direction of an ordinate axis which is the amplitude of said Cepstrum signal and has dimension in the direction of the abscissa which is a quantity representing time, said signal conditioning step including,
• generating a weighting signal which linearly increases in value with the dimension in the direction of the abscissa corresponding to said quantity representing time of said Cepstrum, and
• adding said weighting signal to said Cepstrum signal along said dimension in the direction of the abscissa thereof to generate a weighted Cepstrum signal.

A system for synthesizing speech in real time from a plurality of sets of control parameters representative of a plurality of speech signals, said parameters including a pitch period parameter, a voiced/unvoiced decision parameter, a predetermined number of coefficient parameters and a power parameter, said system comprising:

• pitch pulse generator means for receiving said pitch period parameter and generating pitch pulses having a corresponding pitch period;
• random noise generator means;
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said voiced/unvoiced decision parameter, an output of said switch means being derived from the output of said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced;
• gain control means connected to said output of said switch means, for receiving said power parameter, an output level of said gain control means being dependent upon said power parameter;
• and
• linear prediction filter means connected to said output of said control means for receiving and applying said coefficient parameters to said filter means to generate a replica of each of said plurality of speech signals at an output of said filter means.

A system for synthesizing speech from a combination of control parameters representative of a speech signal, said parameters including a pitch period parameter, a voiced/unvoiced decision parameter, a predetermined number of coefficient parameters and a power parameter, said system comprising:

• pitch pulse generator means for receiving said pitch period parameter and generating pitch pulses having a corresponding pitch period;
• random noise generator means;
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said voiced/unvoiced decision parameter, an output of said switch means being derived from the output of said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced;
• gain control means connected to said output of said switch means, for receiving said power parameter, an output level of said gain control means being dependent upon said power parameter;
• and
• linear prediction filter means connected to said output of said control means, for receiving and applying said coefficient parameters to said filter means to generate a replica of said speech signal at an output of said filter means;
• wherein said pitch pulse generator means includes
• means responsive to receipt of said pitch period parameter for generating a sequence of control signals, and
• means responsive to said sequence of control signals for generating a substantially periodic, amplitude increasing pitch pulse excitation signal.

A method of synthesizing speech in real time from a plurality of combinations of control parameters representative of a plurality of speech signals, said parameter including a pitch period parameter, a voiced/unvoiced decision parameter, a predetermined number of coefficient parameters and a power parameter, said method comprising the steps of:

• receiving said pitch period parameter and generating pitch period pulses having a corresponding pitch period;
• generating a random noise signal;
• switching said pitch period signals to an output of a switch means when said voiced/unvoiced decision parameter is voiced and switching said random noise signal to said output of said switch means if said voiced/unvoiced decision parameter is unvoiced;
• controlling the gain of the output of said switch means in response to said power parameter;
• and
• connecting the output of said gain controlled switch means to a linear prediction filter means connected to said coefficient parameters to generate a replica of each of said plurality of speech signals at an output of said filter means.

A method of synthesizing speech from a combination of control parameters representative of a speech signal, said parameters including a pitch period parameter, a voiced/unvoiced decision parameter, a predetermined number of coefficient parameters and a power parameter, said method comprising the steps of:

• receiving said pitch period parameter and generating pitch period pulses having a corresponding pitch period;
• generating a random noise signal;
• switching said pitch period signals to an output of a switch means when said voiced/unvoiced decision parameter is voiced and switching said random noise signal to said output of said switch means if said voiced/unvoiced decision parameter is unvoiced;
• controlling the gain of the output of said switch means in response to said power parameter;
• and
• connecting the output of said gain controlled switch means to a linear prediction filter means connected to said coefficient parameters to generate a replica of said speech signal at an output of said filter means;
• wherein said pitch pulse generating step includes
• generating a sequence of control signals in response to receipt of said pitch period parameter;
• and
• generating a substantially periodic, amplitude increasing pitch period pulse excitation signal in response to said sequence of control signals.

A system for speech analysis and synthesis, comprising:

• an analyzer including,
• means for analyzing a plurality of speech signal channels in real time by Cepstrum technique to generate a pitch period parameter for each of said speech signal channels,
• means for analyzing each of said plurality of speech signal channels, to generate a voiced/unvoiced decision parameter for each channel,
• means for analyzing each of said plurality of speech signal channels by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter for each of said channels,
• means for combining said pitch period parameter, coefficient parameters, power parameter, and voiced/unvoiced decision parameter into a set of parameters for each of said speech channels,
• means for combining said sets of parameters for each speech channel into a single transmission channel signal,
• means for transmitting said single transmission channel signal to a receiver, and
• means for separating said received single transmission channel signal into respective sets of parameters for a respective plurality of synthesis channels;
• and
• a synthesizer including,
• pitch pulse generator means for receiving said pitch period parameter for each of said synthesis channels and generating pitch pulses having a corresponding pitch period,
• random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said voiced/unvoiced decision parameters for each of said synthesis channels, an output of said switch means being said pitch pulse generator means if said decision for a channel is voiced and said output of said switch means being said random noise generator means if said decision for a channel is unvoiced,
• gain control means connected to said output of said switch means for receiving said power parameters for each of said synthesis channels, an output level of said gain control means being dependent upon said power parameter for each of said synthesis channels, and
• linear prediction filter means connected to said output of said gain control means for receiving and applying said coefficient parameters for each of said synthesis channels to said filter means to generate a replica of each of said speech signals at an output of said filter means.

A system for speech analysis and synthesis, comprising:

• an analyzer including,
• means for analyzing a plurality of speech signal channels by Cepstrum technique to generate a pitch period parameter for each of said speech signal channels,
• means for analyzing each of said plurality of speech signal channels, to generate a voiced/unvoiced decision parameter for each channel,
• means for analyzing each of said plurality of speech signal channels by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter for each of said channels,
• means for combining said pitch period parameter, coefficient parameters, power parameter, and voiced/unvoiced decision parameter into a set of parameters for each of said speech channels,
• means for combining said sets of parameters for each speech channel into a single transmission channel signal,
• means for transmitting said single transmission channel signal to a receiver, and
• means for separating said received single transmission channel signal into respective sets of parameters for a respective plurality of synthesis channels;
• and
• a synthesizer including,
• pitch pulse generator means for receiving said pitch period parameter for each of said synthesis channels and generating pitch pulses having a corresponding pitch period,
• random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said voiced/unvoiced decision parameters for each of said synthesis channels, an output of said switch means being said pitch pulse generator means if said decision for a channel is voiced and said output of said switch means being said random noise generator means if said decision for a channel is unvoiced,
• gain control means connected to said output of said switch means for receiving said power parameters for each of said synthesis channels, an output level of said gain control means being dependent upon said power parameter for each of said synthesis channels, and
• linear prediction filter means connected to said output of said gain control means for receiving and applying said coefficient parameters for each of said synthesis channels to said filter means to generate a replica of each of said speech signals at an output of said filter means;
• wherein said means for analyzing each of said plurality of speech signal channels said speech signal to generate a voiced/unvoiced decision parameter includes
• means for scanning a Cepstrum signal within an auxilliary signal range on each side of a Cepstrum peak amplitude signal,
• means for detecting and storing a range peak signal within the scanned auxilliary ranges,
• means for summing the absolute values of the signal within the auxilliary ranges to form a range signal sum,
• means for dividing said Cepstrum peak amplitude signal by said range peak signal to form a range peak factor,
• means for dividing said range signal sum by said Cepstrum peak amplitude signal to form an average peak factor,
• means for first comparing said range peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said range peak factor is above said upper limit and generating an unvoiced decision parameter if said range peak factor is equal to or below said lower limit, and
• means for secondarily comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A system for speech analysis and synthesis, comprising:

• an analyzer including,
• means for analyzing a plurality of speech signal channels by Cepstrum technique to generate a pitch period parameter for each of said speech signal channels,
• means for analyzing each of said plurality of speech signal channels, to generate a voiced/unvoiced decision parameter for each channel,
• means for analyzing each of said 1lurality of speech signal channels by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter for each of said channels,
• means for combining said pitch period parameter, coefficient parameters, power parameter, and voiced/unvoiced decision parameter into a set of parameters for each of said speech channels,
• means for combining said sets of parameters for each speech channel into a single transmission channel signal,
• means for transmitting said single transmission channel signal to a receiver, and
• means for separating said received single transmission channel signal into respective sets of parameters for a respective plurality of synthesis channels;
• and
• a synthesizer including,
• pitch pulse generator means for receiving said pitch period parameter for each of said synthesis channels and generating pitch pulses having a corresponding pitch period,
• random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said voiced/unvoiced decision parameters for each of said synthesis channels, an output of said switch means being said pitch pulse generator means if said decision for a channel is voiced and said output of said switch means being said random noise generator means if said decision for a channel is unvoiced,
• gain control means connected to said output of said switch means for receiving said power parameters for each of said synthesis channels, an output level of said gain control means being dependent upon said power parameter for each of said synthesis channels, and
• linear prediction filter means connected to said output of said gain control means for receiving and applying said coefficient parameters for each of said synthesis channels to said filter means to generate a replica of each of said speech signals at an output of said filter means;
• wherein said means for analyzing the plurality of speech signal channels by Cepstrum technique to generate a pitch period parameter includes a signal conditioning system for conditioning a Cepstrum signal which has, in the direction of an ordinate axis, an amplitude dimension of said Cepstrum signal and, in the direction of the abscissa, a quantity representing time, said signal conditioning system having,
• means for generating a weighting signal which linearly increases in value with the dimension in the direction of the abscissa corresponding to said quantity representing time of said Cepstrum, and
• means for adding said weighting signal to said Cepstrum signal along said dimension in the direction of the abscissa thereof to generate a weighted Cepstrum signal.

A system for analyzing a plurality of speech signal channels, in real time, said system comprising an analyzer including:

• means for analyzing a plurality of speech signal channels by Cepstrum technique to generate a pitch period parameter for each of said speech signal channels;
• means for analyzing each of said plurality of speech signal channels to generate a voiced/unvoiced decision parameter for each channel;
• and
• means for analyzing each of said plurality of speech signal channels by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter for each of said channels.

A system for analyzing a plurality of speech signal channels, comprising an analyzer including:

• means for analyzing a plurality of speech signal channels by Cepstrum technique to generate a pitch period parameter for each of said speech signal channels;
• means for analyzing each of said plurality of speech signal channels to generate a voiced/unvoiced decision parameter for each channel;
• and
• means for analyzing each of said plurality of speech signal channels by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter for each of said channels;
• wherein said means for analyzing said speech signal to generate a voiced/unvoiced decision parameter includes
• means for scanning a Cepstrum signal within an auxilliary signal range on each side of Cepstrum peak amplitude signal,
• means for detecting and storing a range peak signal within the scanned auxilliary ranges,
• means for summing the absolute values of the signal within the auxilliary ranges to form a range signal sum,
• means for dividing said Cepstrum peak amplitude signal by said range peak signal to form a range peak factor;
• means for dividing said range signal sum by said Cepstrum peak amplitude signal to form an average peak factor,
• means for first comparing said range peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said range peak factor is above said upper limit and generating an unvoiced decision parameter if said range peak factor is equal to or below said lower limit, and
• means for secondarily comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A system for analyzing a plurality of speech signal channels, comprising an analyzer including:

• means for analyzing a plurality of speech signal channels by Cepstrum technique to generate a pitch period parameter for each of said speech signal channels;
• means for analyzing each of said plurality of speech signal channels to generate a voiced/unvoiced decision parameter for each channel;
• and
• means for analyzing each of said plurality of speech signal channels by linear prediction technique to generate a predetermined number of coefficient parameters and a power parameter for each of said channels;
• wherein said means for analyzing a plurality of speech signal channels by Cepstrum technique includes a signal conditioning system for conditioning a Cepstrum signal which has, in the direction of an ordinate axis, an amplitude dimension of said Cepstrum signal and, in the direction of the abscissa, a quantity representing time, said signal conditioning system having,
• means for generating a weighting signal which linearly increases in value with the dimension in the direction of the abscissa corresponding to said quantity representing time of said Cepstrum, and
• means for adding said weighting signal to said Cepstrum signal along said dimension in the direction of the abscissa thereof to generate a weighted Cepstrum signal.

A system for synthesizing speech in a plurality of speech synthesis channels in real time, said system comprising:

• pitch pulse generator means for receiving pitch period parameters for each of said synthesis channels and generating pitch pulses having a corresponding pitch period;
• random noise generator means;
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving voiced/unvoiced decision parameters for each of said synthesis channels, an output of said switch means being derived from said pitch pulse generator means if said decision for a channel is voiced and from said random noise generator means if said decision for a channel is unvoiced;
• gain control means connected to said output of said switch means for receiving power parameters for each of said synthesis channels, an output level of said gain control means being dependent upon said power parameter for each of said synthesis channels;
• and
• linear prediction filter means connected to said output of said gain control means for receiving and applying a plurality of coefficient parameters for each of said synthesis channels to said filter means to generate a replica of each of said speech signals at an output of said filter means.

A system for speech synthesis, comprising a synthesizer including:

• pitch pulse generator means for receiving pitch period parameters for each of said synthesis channels and generating pitch pulses having a corresponding pitch period;
• random noise generator means;
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving voiced/unvoiced decision parameters for each of said synthesis channels, an output of said switch means being derived from said pitch pulse generator means if said decision for a channel is voiced and from said random noise generator means if said decision for a channel is unvoiced;
• gain control means connected to said output of said switch means for receiving power parameters for each of said synthesis channels, an output level of said gain control means being dependent upon said power parameter for each of said synthesis channels;
• and
• linear prediction filter means connected to said output of said gain control means for receiving and applying a plurality of coefficient parameters for each of said synthesis channels to said filter means to generate a replica of each of said speech signals at an output of said filter means;
• wherein said pitch pulse generator means includes
• means responsive to receipt of said pitch period parameters for generating a sequence of control signals, and
• means responsive to said sequence of control signals for generating a substantially periodic, amplitude increasing pitch pulse excitation signal.

A speech analysis system for generating in real time a plurality of sets of paramaters each representative of an analog speech signal as produced by a model speech generator, said system comprising:

• means for sampling each of said analog speech signals to produce a plurality of digital speech data signals;
• means for separating a fixed number of said speech data signals into a frame of speech data signals;
• means for analyzing said frame of said speech data with a Cepstrum generator to produce a digital speech pitch period parameter for said frame of data;
• means for analyzing said frame of data with a linear prediction coefficient generator to produce a plurality of digital coefficeint generator to produce a plurality of digital coefficeint parameters and a digital power parameter;
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter;
• and
• means for combining, for each speech signal, said digital pitch period parameter, said plurality of digital coefficient parameters, said digital power parameter, and said digital voiced/unvoiced decision parameter into a set of parameters representative of each of said analog speech signals.

A speech analysis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator, said system comprising:

• means for sampling said analog speech signal to produce a plurality of digital speech data signals;
• means for separating a fixed number of said speech data signals into a frame of speech data signals;
• means for analyzing said frame of said speech data with a Cepstrum generator to produce a digital speech pitch period parameter for said frame of data;
• means for analyzing said frame of data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter;
• and
• means for combining said digital pitch period parameter, said plurality of digital coefficient parameters, said digital power parameter, and said digital voiced/unvoiced decision paramater into said set of parameters representative of said analog speech signal;
• wherein said means for analyzing said frame of speech data to produce a digital voiced/unvoiced decision parameter includes
• means for scanning a Cepstrum signal within an auxilliary signal range on each side of a Cepstrum peak amplitude signal,
• means for detecting and storing a range peak signal within the scanned auxilliary ranges,
• means for summing the absolute values of the signal within the auxilliary ranges to form a range signal sum,
• means for dividing said Cepstrum peak amplitude signal by said range peak signal to form a range peak factor,
• means for dividing said range signal sum by said Cepstrum peak amplitude signal to form an average peak factor,
• means for first comparing said range peak factor will predetermined upper and lower limits and generating a voiced decision parameter if said range peak factor is above said upper limit and generating an unvoiced decision parameter if said range peak factor is equal to or below said lower limit, and
• means for secondarily comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A speech analysis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator, said system comprising:

• means for sampling said analog speech signal to produce a plurality of digital speech data signals;
• means for separating a fixed number of said speech data signals into a frame of speech data signals;
• means for analyzing said frame of said speech data with a Cepstrum generator to produce a digital speech pitch period parameter for said frame of data;
• means for analyzing said frame of data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter;
• and
• means for combining said digital pitch period parameter, said plurality of digital coefficient parameters, said digital power parameter, and said digital voiced/unvoiced decision parameter into said set of parameters representative of said analog speech signal;
• wherein said means for analyzing said frame of speech data by Cepstrum technique includes a signal conditioning system for conditioning a Cepstrum signal which has, in the direction of an ordinate axis, an amplitude dimension of said Cepstrum signal and, in the direction of the abscissa, a quantity representing time, said signal conditioning system having,
• means for generating a weighting signal which linearly increases in value with the dimension in the direction of the abscissa corresponding to said quantity representing time of said Cepstrum, and
• means for adding said weighting signal to said Cepstrum signal along said dimension in the direction of the abscissa thereof to generate a weighted Cepstrum signal.

A speech analysis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator, said system comprising:

• means for sampling said analog speech signal to produce a plurality of digital speech data signals;
• means for separating a fixed number of said speech data signals into a frame of speech data signals;
• means for analyzing said frame of said speech data with a Cepstrum generator to produce a digital speech pitch period parameter for said frame of data;
• means for analyzing said frame of data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter;
• and
• means for combining said digital pitch period parameter, said plurality of digital coefficient parameters, said digital power parameter, and said digital voiced/unvoiced decision parameter into said set of parameters representative of said analog speech signal;
• wherein said means for analyzing said frame of said speech data to produce a speech pitch period parameter includes means for interpolating between sample values in a sample data Cepstrum signal, and wherein a pitch period value is initially chosen as corresponding to a maximum amplitude sample, said interpolating system having
• means for storing said maximum value sample, an immediately preceding sample, and an immediately following sample,
• means for subtracting said immediately preceding sample from said immediately following sample to form a sample difference,
• means for dividing said sample difference by four to form a sample difference quotient,
• means for adding said immediately preceding sample to said immediately following sample to form a sample sum,
• means for dividing said sample sum by two to form a sample sum quotient,
• means for subtracting said maximum amplitude sample from said sample sum quotient to form a weighted sample sum quotient,
• means for dividing said sample difference quotient by said weighted sample sum quotient to form a pitch value increment, and
• means for adding said pitch value increment to said pitch period value to form an interpolated pitch value.

A speech analysis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator, said system comprising:

• means for sampling said analog speech signal to produce a plurality of digital speech data signals;
• means for separating a fixed number of said speech data signals into a frame of speech data signals;
• means for analyzing said frame of said speech data with a Cepstrum generator to produce a digital speech pitch period parameter for said frame of data;
• means for analyzing said frame of data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter;
• and
• means for combining said digital pitch period parameter, said plurality of digital coefficient parameters, said digital power parameter, and said digital voiced/unvoiced decision parameter into said set of parameters representative of said analog speech signal;
• wherein said means for analyzing said frame of data with a linear prediction coefficient generator includes a signal conditioning means for a frame of digital speech data samples, said signal conditioning means having
• means for generating a window signal having a characteristic waveshape which is completed in a data sample span which is shorter than said frame of data;
• means for centering said window signal data span in said frame of speech data samples;
• and
• means for multiplying corresponding sample positions of said frame of speech data and said window signal to generate a limited and windowed frame of speech data.

A speech analysis and synthesis system for generating a plurality of sets of parameters representative of a plurality of analog speech signals as produced by a model speech generator and reproducing in real time from said sets of parameters a replica of each of said analog speech signals, said system comprising:

• an analyzer including,
• means for sampling each analog speech signal to produce a plurality of digital speech data signals,
• means for separating said plurality of said speech data signals into a frame of speech data signals,
• means for analyzing said frame of speech data with a Cepstrum generator to produce a digital pitch period parameter for said frame of data,
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter, and
• means for analyzing said frame of speech data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• and
• a synthesizer including,
• digital pitch pulse generator means for receiving said digital pitch period parameter and generating for each original speech signal pitch pulses having a corresponding pitch period,
• digital random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said digital voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision in unvoiced,
• gain control means connected to said output of said switch means for receiving said digital power parameter, a digital output level of said gain control means being dependent upon said power parameter,
• digital linear prediction filter means connected to said output of said gain control means for receiving and applying said digital coefficient parameters to said filter means to generate a digital replica of each of said original speech signals at an output of said filter means, and
• means for converting said digital output of said filter means to a corresponding analog replica of each of said original analog speech signals.

A speech analysis and synthesis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator and reproducing from said set of parameters a replica of said analog speech signal, said system comprising:

• an analyzer including,
• means for sampling said analog speech signal to produce a plurality of digital speech data signals,
• means for separating said plurality of said speech data signals into a frame of speech data signals,
• means for analyzing said frame of speech data with a Cepstrum generator to produce a digital pitch period parameter for said frame of data,
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter, and
• means for analyzing said frame of speech data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• and
• a synthesizer including,
• digital pitch pulse generator means for receiving said digital pitch period parameter and generating pitch pulses having a corresponding pitch period,
• digital random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said digital voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced,
• gain control means connected to said output of said switch means for receiving said digital power parameter, a digital output level of said gain control means being dependent upon said power parameter,
• digital linear prediction filter means connected to said output of said gain control means for receiving and applying said digital coefficient parameters to said filter means to generate a digital replica of said speech signal at an output of said filter means, and
• means for converting said digital output of said filter means to a corresponding analog replica of said analog speech signal;
• wherein said means for analyzing said frame of speech data to produce a digital voiced/unvoiced decision parameter includes
• means for scanning a Cepstrum signal within an auxilliary signal range on each side of a Cepstrum peak amplitude signal,
• means for detecting and storing a range peak signal within the scanned auxilliary ranges,
• means for summing the absolute values of the signal within the auxilliary ranges to form a range signal sum,
• means for dividing said Cepstrum peak amplitude signal by said range peak signal to form a range peak factor;
• means for dividing said range signal sum by said Cepstrum peak amplitude signal to form an average peak factor,
• means for first comparing said range peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said range peak factor is equal to or above said upper limit and generating an unvoiced decision if said range peak factor is equal to or below said lower limit, and
• means for secondarily comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A speech analysis and synthesis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator and reproducing from said set of parameters a replica of said analog speech signal, said system comprising:

• an analyzer including,
• means for sampling said analog speech signal to produce a plurality of digital speech data signals,
• means for separating said plurality of said speech data signals into a frame of speech data signals,
• means for analyzing said frame of speech data with a Cepstrum generator to produce a digital pitch period parameter for said frame of data,
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter, and
• means for analyzing said frame of speech data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• and
• a synthesizer including,
• digital pitch pulse generator means for receiving said digital pitch period parameter and generating pitch pulses having a corresponding pitch period,
• digital random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said ditigal voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced,
• gain control means connected to said output of said switch means for receiving said digital power parameter, a digital output level of said gain control means being dependent upon said power parameter,
• digital linear prediction filter means connected to said output of said gain control means for receiving and applying said ditigal coefficient parameters to said filter means to generate a digital replica of said speech signal at an output of said filter means, and
• means for converting said digital output of said filter means to a corresponding analog replica of said analog speech signal;
• wherein said means for analyzing said frame of speech data with a Cepstrum generator includes a signal conditioning system for conditioning a Cepstrum signal which has, in the direction of an ordinate axis, an amplitude dimension of said Cepstrum signal and, in the direction of the abscissa, a quantity representing time, said signal conditioning system having,
• means for generating a weighting signal which linearly increases in value with the dimension in the direction of the abscissa corresponding to said quantity representing time of said Cepstrum, and
• means for adding said weighting signal to said Cepstrum signal along said dimension in the direction of the abscissa thereof to generate a weighted Cepstrum signal.

A speech analysis and synthesis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator and reproducing from said set of parameters a replica of said analog speech signal, said system comprising:

• an analyzer including,
• means for sampling said analog speech signal to produce a plurality of digital speech data signals,
• means for separating said plurality of said speech data signals into a frame of speech data signals,
• means for analyzing said frame of speech data with a Cepstrum generator to produce a digital pitch period parameter for said frame of data,
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter, and
• means for analyzing said frame of speech data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• and
• a synthesizer including,
• digital pitch pulse generator means for receiving said digital pitch period parameter and generating pitch pulses having a corresponding pitch period,
• digital random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said digital voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced,
• gain control means connected to said output of said switch means for receiving said digital power parameter, a digital output level of said gain control means being dependent upon said power parameter,
• digital linear prediction filter means connected to said output of said gain control means for receiving and applying said digital coefficient parameters to said filter means to generate a digital replica of said speech signal at an output of said filter means, and
• means for converting said digital output of said filter means to a corresponding analog replica of said analog speech signal;
• wherein said digital pitch pulse generator means comprises
• means responsive to receipt of said digital pitch period parameter, for generating a sequence of control signals, and
• means responsive to said sequence of control signals for generating a substantially periodic, amplitude increasing pitch pulse excitation signal.

A speech analysis and synthesis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator and reproducing from said set of parameters a replica of said analog speech signal, said system comprising:

• an analyzer including,
• means for sampling said analog speech signal to produce a plurality of digital speech data signals,
• means for separating said plurality of said speech data signals into a frame of speech data signals,
• means for analyzing said frame of speech data with a Cepstrum generator to produce a digital pitch period parameter for said frame of data,
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter, and
• means for analyzing said frame of speech data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• and
• a synthesizer including,
• digital pitch pulse generator means for receiving said digital pitch period parameter and generating pitch pulses having a corresponding pitch period,
• digital random noise generator means,
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said digital voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced,
• gain control means connected to said output of said switch means for receiving said digital power parameter, a digital output level of said gain control means being dependent upon said power parameter,
• digital linear prediction filter means connected to said output of said gain control means for receiving and applying said digital coefficient parameters to said filter means to generate a digital replica of said speech signal at an output of said filter means, and
• means for converting said digital output of said filter means to a corresponding analog replica of said analog speech signal;
• wherein said means for analyzing said frame of said speech data to produce a digital pitch period parameter includes means for interpolating between sample values in a sample data Cepstrum signal wherein a pitch period value is initially chosen as corresponding to a maximum amplitude sample, said interpolating system having
• means for storing said maximum value sample, an immediately preceding sample, and an immediately following sample,
• means for substracting said immediately preceding sample from said immediately following sample to form a sample difference,
• means for dividing said sample difference by four to form a sample difference quotient,
• means for adding said immediately preceding sample to said immediately following sample to form a sample sum,
• means for dividing said sample sum by two to form a sample sum quotient,
• means for subtracting said maximum amplitude sample from said sample sum quotient to form a weighted sample sum quotient,
• means for dividing said sample difference quotient by said weighted sample sum quotient to form a pitch value increment, and
• means for adding said pitch value increment to said pitch period value to form an interpolated pitch value.

A speech analysis and synthesis system for generating a set of parameters representative of an analog speech signal as produced by a model speech generator and reproducing from said set of parameters a replica of said analog speech signal, said system comprising:

• an analyzer including,
• means for sampling said analog speech signal to produce a plurality of digital speech data signals,
• means for separating said plurality of said speech data signals into a frame of speech data signals,
• means for analyzing said frame of speech data with a Cepstrum generator to produce a digital pitch period parameter for said frame of data,
• means for analyzing said frame of speech data with a voiced/unvoiced decision generator to produce a digital voiced/unvoiced decision parameter, and
• means for analyzing said frame of speech data with a linear prediction coefficient generator to produce a plurality of digital coefficient parameters and a digital power parameter;
• and
• a synthesizer including,
• digital pitch pulse generator means for receiving said digital pitch period parameter and generating pitch pulses having a corresponding pitch period,
• digital random noise generator means,
• switch means connected to said pitch pulses generator means and to said random noise generator means for receiving said digital voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced,
• gain control means connected to said output of said switch means for receiving said digital power parameter, a digital output level of said gain control means being dependent upon said power parameter,
• digital linear prediction filter means connected to said output of said gain control means for receiving and applying said digital coefficient parameters to said filter means to generate a digital replica of said speech signal at an output of said filter means, and
• means for converting said digital output of said filter means to a corresponding analog replica of said analog speech signal;
• wherein said means for analyzing said frame of data with a linear prediction coefficient generator includes a signal conditioning means for a frame of digital speech data samples, said signal conditioning means having
• means for generating a window signal having a characteristic waveshape which is completed in a data sample span which is shorter than said frame of data,
• means for centering said window signal data sample span in said frame of speech data samples, and
• means for multiplying corresponding sample positions of said frame of speech data and said window signal to generate a limited and windowed frame of speech data.

A speech synthesis system for generating in real time an analog replica of each of a plurality of original analog speech signals represented by corresponding sets of digital parameters, each set of said parameters including a digital speech pitch period parameter, a digital voiced/unvoiced decision parameter, a plurality of digital coefficient parameters and a digital power parameter, said synthesis system including:

• digital pitch pulse generator means for receiving said digital pitch period parameter and generating pitch pulses having a corresponding pitch period;
• digital random noise generator means;
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said digital voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced;
• gain control means connected to said output of said switch means for receiving said digital power parameter, a digital output level of said gain control means being dependent upon said power parameter;
• digital linear prediction filter means connected to said output of said gain control means for receiving and applying said digital coefficient parameters to said filter means to generate a digital replica of each of said plurality of speech signals at an output of said filter means;
• and
• means for converting said digital output of said filter means to a corresponding analog replica of each of said plurality of analog speech signals.

A speech synthesis system for generating an analog replica of an original analog speech signal represented by a set of digital parameters, said parameters including a digital speech pitch period parameter, a digital voiced/unvoiced decision parameter, a plurality of digital coefficient parameters and a digital power parameter, said synthesis system including:

• digital pitch pulse generator means for receiving said digital pitch period parameter and generating pitch pulses having a corresponding pitch period;
• digital random noise generator means;
• switch means connected to said pitch pulse generator means and to said random noise generator means for receiving said digital voiced/unvoiced decision parameter, an output of said switch means being derived from said pitch pulse generator means if said decision is voiced and from said random noise generator means if said decision is unvoiced;
• gain control means connected to said output of said switch means for receiving said digital power parameter, a digital output level of said gain control means being dependent upon said power parameter;
• digital linear prediction filter means connected to said output of said gain control means for receiving and applying said digital coefficient parameters to said filter means to generate a digital replica of said speech signal at an output of said filter means;
• and
• means for converting said digital output of said filter means to a corresponding analog replica of said analog speech signal;
• wherein said digital pitch pulse generator means includes
• means responsive to receipt of said digital pitch period parameter, for generating a sequence of control signals, and
• means responsive to said sequence of control signals for generating a substantially periodic, amplitude increasing pitch pulse excitation signal.

A system for generating a voiced/unvoiced decision parameter in a speech analysis system wherein a pitch parameter is generated by a Cepstrum technique and coefficient parameters are generated by a linear prediction technique, said system comprising:

• means for scanning said Cepstrum signal within an auxilliary sample range on each side of a Cepstrum peak amplitude sample;
• means for detecting and storing a range peak sample within the scanned auxilliary ranges;
• means for summing the absolute values of the samples within the auxilliary ranges to form a range sample sum;
• means for dividing said Cepstrum peak amplitude sample by said range peak sample to form a range peak factor;
• means for dividing said range sample sum by said Cepstrum peak amplitude sample to form an average peak factor;
• means for first comparing said range peak factor with predetermined upper and lower limits and generating an unvoiced decision parameter if said range peak factor is above said upper limit and generating an unvoiced decision parameter if said range peak factor is equal to or below said lower limit;
• and
• means for secondly comparing said average peak factor with predetermined upper and lower limits and generating a voiced decision parameter if said average peak factor is equal to or less than said lower limit and generating an unvoiced decision parameter if said average peak factor is equal to or greater than said upper limit.

A signal conditioning system for a frame of digital speech data samples, said system comprising:

• means for generating a window signal having a characteristic waveshape which is completed in a data sample span which is shorter than said frame of data;
• means for centering said window signal data sample span in said frame of speech data samples;
• and
• means for multiplying corresponding sample positions of said frame of speech data and said window signal to generate a limited and windowed frame of speech data.