Method and apparatus for identifying speech in telephone signals

A method for identifying a signal which is produced by a human voice which comprises the steps of:

• (1) determining when the signal exceeds a first predetermined level;
• (2) obtaining the amplitude of the signal and providing a digital representation of the amplitude, if the signal is a digital signal, or obtaining a sample of the amplitude of the signal and providing a digital representation of the sample, if the signal is an analog signal, and grouping the digital representations into frames, a frame being comprised of a predetermined number of consecutive digital representations of the amplitude of the signal, if the signal is a digital signal, or digital representations of samples of the amplitude of the signal, if the signal is an analog signal;
• (3) determining the contents and a measure of the energy of predetermined frequency bins from the contents of the frequency spectrum for each frame;
• (4) summing the measure of the energy of each of the predetermined frequency bins for each frame and over a predetermined multiplicity of frames, a frame group, to provide frame group energy measures;
• (5) determining when the signal falls below a second predetermined level;
• (6) determining the time period between the time the signal exceeded the first predetermined level and the time the signal fell below the second predetermined amplitude and identifying the signal as not having been produced by a human voice if the time period is shorter than a first predetermined time period;
• (7) determining a sum of the absolute value of the difference between the frame group energy measures for all pairs of frame groups which are consecutive in time during the time period and comparing the sum with a predetermined amount;
• and
• (8) identifying the signal as having been produced by a human voice if the sum exceeds the predetermined amount and identifying the signal as not having been produced by a human voice if the sum does not exceed the predetermined amount.

A method for identifying a signal which is produced by a human voice speech which comprises the steps of:

• (1) determining when the signal exceeds a first predetermined level;
• (2) obtaining the amplitude of the signal and providing a digital representation of the amplitude, if the signal is a digital signal, or obtaining a sample of the amplitude of the signal and providing a digital representation of the sample, if the signal is an analog signal, and grouping the digital representations into frames, a frame being comprised of a predetermined number of consecutive digital representation of the amplitude of the signal, if the signal is a digital signal, or digital representations of samples of the amplitude of the signal, if the signal is an analog signal;
• (3) determining the contents and a measure of the energy of predetermined frequency bins from the contents of the frequency spectrum of the signal for each frame;
• (4) determining the sum of the measure of the energy of the predetermined frequency bins and a frequency weighted sum of the measure of the energy of the predetermined frequency bins for each frame and over a predetermined multiplicity of frames, a frame group, to provide frame group energy measures and frame group frequency weighted energy measures;
• (5) determining when the signal falls below a second predetermined level;
• (6) determining the time period between the time the signal exceeded the first predetermined level and the time the signal fell below the second predetermined amplitude and identifying the signal as not having been produced by a human voice if the time period is shorter than a first predetermined time period;
• (7) determining a first sum of the absolute value of the difference between the frame group energy measure for all pairs of frame groups which are consecutive in time during the time period and determining a second sum of the absolute value of the difference between the frame group frequency weighted energy measure for all pairs of frame groups which are consecutive in time during the time period;
• and
• (8) comparing the first sum with a first predetermined amount and comparing the second sum with a second predetermined amount and identifying the signal as having been produced by a human voice if the first sum exceeds the first predetermined amount and the second sum exceeds the second predetermined amount.

An apparatus for identifying a signal which is produced by a human voice speech which comprises:

• means for detecting when the signal exceeds a first predetermined level and, in response thereto, for activating means for obtaining;
• the means for obtaining being means for obtaining the amplitude of the signal and for providing a digital representation of the amplitude, if the signal is a digital signal, or for obtaining a sample of the amplitude of the signal and for providing a digital representation of the sample, if the signal is an analog signal, and for grouping the digital representations into frames, a frame being comprised of a predetermined number of consecutive digital representation of the amplitude of the signal, if the signal is a digital signal, or digital representations of samples of the amplitude of the signal, if the signal is an analog signal;
• means for determining the contents and a measure of the energy of predetermined frequency bins from the contents of the frequency spectrum of the signal for each frame;
• means for summing the measure of the energy of each of the predetermined frequency bins for each frame over a predetermined multiplicity of frames, a frame group, to provide a set of frame group energy measures for the predetermined frequency bins;
• means for detecting when the signal falls below a second predetermined level and, in response thereto, activating means for determining a time period;
• the means for determining being means for determining the time period between the time the signal exceeded the first predetermined level and the time the signal fell below the second predetermined level and means for identifying the signal as not having been produced by a human voice if the time period is shorter than a first predetermined time period;
• and
• means for comparing the variation in the set of frame group energy measures for all pairs of frame groups which are consecutive in time during the time period with a predetermined library of variations and for identifying the signal as having been produced by a human voice if the step of comparing finds a match with one of the library.

An apparatus for identifying a signal which is produced by a human voice speech which comprises:

• means for detecting when the signal exceeds a first predetermined level and, in response thereto, for activating means for obtaining;
• the means for obtaining being means for obtaining the amplitude of the signal and for providing a digital representation of the amplitude, if the signal is a digital signal, or for obtaining a sample of the amplitude of the signal and for providing a digital representation of the sample, if the signal is an analog signal, and for grouping the digital representations into frames, a frame being comprised of a predetermined number of consecutive digital representation of the amplitude of the signal, if the signal is a digital signal, or digital representations of samples of the amplitude of the signal, if the signal is an analog signal;
• means for determining the contents and a measure of the energy of the predetermined frequency bins from the contents of the frequency spectrum of the signal for each frame;
• means for determining the sum of the measure of the energy of the predetermined frequency bins and a frequency weighted sum of the measure of the energy of the predetermined frequency bins for each frame and over a predetermined multiplicity of frames, a frame group, to provide frame group energy measures and frame group frequency weighted energy measures;
• means for detecting when the signal falls below a second predetermined level and, in response thereto, activating means for determining a time period;
• the means for determining being means for determining the time period between the time the signal exceeded the first predetermined level and the time the signal fell below the second predetermined level and means for identifying the signal as not having been produced by a human voice if the time period is shorter than a first predetermined time period;
• means for determining a first sum of the absolute value of the difference between the frame group energy measure for all pairs of frame groups which are consecutive in time during the time period and determining a second sum of the absolute value of the difference between the frame group frequency weighted energy measure for all pairs of frame groups which are consecutive in time during the time period;
• and
• means for comparing the first sum with a first predetermined amount and comparing the second sum with a second predetermined amount and for identifying the signal as having been produced by a human voice if the first sum exceeds the first predetermined amount and the second sum exceeds the second predetermined amount.