Spectrogram-feature-based speech syllable and word recognition using syllabic language dictionary

Claim 2

A speech deciphering device comprising: (a) a digital device having an input processing element, a central processing element, a data storage element, an output processing element and an analog to digital converting element; (b) therein being present logic in the digital device, for (1) segmenting in time and frequency speech syllables, (2) recognizing the speech syllable by its constituting components of vocal region, root syllable, vowel variation and syllable duration, (3) represent for storage the individual speech syllable by its constituting components used in syllable recognition, and (4) deciphering speech made up of the time sequence of speech syllables by mapping a syllable sequence arranged speech word to a known language word using a syllablically arranged dictionary.

Claim 3

The speech deciphering device of claim 1, wherein, the speech syllable segmenting in time and frequency logic comprising: (a) input to the analog to digital converting element of an amplitude versus time speech signal, and the output being a digital speech signal; (b) input processing element storing the digital speech signal into the data storage element; (c) the computer processing element repeatable reading a sequence of the digital speech signal from the data storage element, and performing digital fourier transformation on successive segments of the digital signal, where the transform output data stored in an array in the data storage element, with each array cell location representing the frequency and time, and each cell magnitude representing the sound energy amplitude at that frequency and at that time; (d) the transform output data being normalized in frequency and magnitude; (e) the central processing element processing the normalized data to partition it into Syllable Energy Patches (SEP), where SEP is an area covered by an intersection of a time band and a frequency band, start and end of a band being calculated on the relative magnitude of the sound energy exceeding a threshold; (f) the central processing element computing a speech syllable recognition signature based on the SEP's time and frequency identification and peaks of the energy in the SEP, a peak being defined as any cell location, in which the sound energy is greater than the sound energy in its surrounding cells.

Claim 4

The speech deciphering device of claim 2, wherein, the speech syllable recognition logic comprising: (a) the central processing element comparing the speech syllable recognition signature with a know speech syllable recognition characteristic (SSRC) in recognizing the speech syllable; (b) SSRC being a distinguishing pattern for each syllable wherein SSRC is a function of SEP's location in time and frequency, of the number of peaks, each peak's slopes along the time axis and along the frequency axis, spatial relationship of the peaks, and magnitude of the peaks.

Claim 5

The speech deciphering device of claim 3, wherein, the speech syllable storage logic comprising; (a) storing the speech syllable by a numerical code in a digital word; (b) the numerical code having individual sub-codes for root speech syllable, region of the vocal system speech syllable emanated from, vowel variation of the root speech syllable, and duration of the speech syllable.

Claim 6

Speech storage logic as in claim 4, wherein the numerical code further comprising: (a) a sub-code for gender based sound syllables; (b) a sub-code for language class.

Claim 7

Speech syllable storage logic as in claim 5, wherein the digital word for numerical code arranged as a 16 bit data word.

Claim 8

Speech storage logic as in claim 4, wherein, a string of ordered syllables is stored as a word, a string of ordered words is stored as a sentence, and a string of ordered sentences is stored as a discourse.

Claim 9

The speech deciphering device of claim 4, wherein, the speech deciphering logic comprising: (a) the data storage element having a syllablically arranged dictionary of a language, in which the language word are sorted by their speech syllables and the language words are encoded by their grammatical structure attribute in the language; (b) the central processing element mapping a syllable sequence arranged speech word to a known language word using the syllablically arranged dictionary, leading to language word and its grammatical structure decipher.

Claim 10

The speech deciphering device as in claim 1, further comprising: a device component generating a speech syllable equivalent to the recognized speech syllable.

Claim 11

The device as in claim 9, wherein the speech syllable generation component comprising: (a) the data storage element having stored the codes representing the ensemble of spoken speech syllables; (b) an articulating element which converts the codes to a spoken form; (c) an interface to the digital device, over which commands are received to transfer sequence of speech syllables codes from the data storage element to the articulating element.

Claim 12

A device as in claim 10, wherein the interface being a programmable sequence in time of command codes representing speech syllables of speech words and sentences.

Claim 13

A speech deciphering method, comprising the steps of: (a) using a speech data output in the form of an array of numbers which represent the result of computation of overlapping DFT computation of digitized time domain speech signal; (b) segmenting the array output in time, wherein the segmentation composed of time bands and inter time band gaps based on the variation in magnitude of energy; (c) segmenting the array output in frequency, wherein the segmentation composed of one of seven frequency bands corresponding to the vocal regions, based on magnitude of energy; (d) locating a Syllable Energy Patch (SEP) as an energy pattern over an area that overlaps the intersection of the time band and the frequency band, such that the SEP corresponding to an energy pattern representing a speech syllable; (e) deriving a speech syllable recognition characteristic (SSRC) function for each SEP, wherein the SSRC is a function of SEP's start and end frequencies, start and end time interval, number of peaks, for each peak, its magnitude, location, ascending and descending slope in time and frequency, spatial relationship of peaks via their diagonal distances.

Claim 14

A speech deciphering method, comprising the steps of: (a) using a speech data output in the form of an array of numbers which represent the result of computation of overlapping DFT computation of digitized time domain speech signal; (b) segmenting the array output in time, wherein the segmentation composed of time bands and inter time band gaps based on the variation in magnitude of energy; (c) segmenting the array output in frequency, wherein the segmentation composed of one of seven frequency bands corresponding to the vocal regions, based on magnitude of energy; (d) locating a Syllable Energy Patch (SEP) as an energy pattern over an area that overlaps the intersection of the time band and the frequency band, such that the SEP corresponding to an energy pattern representing a speech syllable; (e) deriving a speech syllable recognition characteristic (SSRC) function for each SEP, wherein the SSRC is a function of SEP's start and end frequencies, start and end time interval, number of peaks, for each peak, its magnitude, location, ascending and descending slope in time and frequency, spatial relationship of peaks via their diagonal distances.

Claim 15

The speech deciphering method, as in claim 12, further comprising the step of: (a) mapping the SSRC into a table of syllable identification codes, wherein each code comprised of sub-codes corresponding to one of set of possible alternatives composed of vocal region, root syllables, root syllable's vowel modulation, and length caused variation.

Claim 16

The speech deciphering method, as in claim 13, wherein, the sub-codes further comprising the step of: language class, based on identification of language class by histogram of occurrence of syllables.

Claim 17

The speech deciphering method, as in claim 14, further comprising the step of: a sub-code comprising gender class based on shift in frequency spectrum required to normalize in frequency.

Claim 18

The speech deciphering method, as in claim 15, further comprising the steps of: making a word string from a group of syllables separated by a inter word gap.

Claim 19

The speech deciphering method, as in claim 16, further comprising the steps of: (a) having a syllablically arranged dictionary of a language, in which the language word are sorted by their spoken speech syllables; (b) mapping the word composed of syllable identification codes into dictionary words, thereby identifying the speech word.

Claim 20

The speech deciphering method, as in claim 17, further comprising the steps of: where the dictionary word is encoded with a sentence structure grammar attribute.

Claim 21

A speech syllable based dictionary useful in deciphering speech using digital computing device, comprising: (a) a database with at least two fields; (b) the first field storing a sequence of speech syllable codes of which a language word is composed such that these syllable based language words are sorted in the order of their syllables from speech pronunciation; (c) the second field storing equivalent language words such that the database can be searched, based on the syllable based language word, and its equivalent language word found.

Claim 22

A speech syllable based dictionary useful in deciphering speech using digital computing device, comprising: (a) a database with at least two fields; (b) the first field storing a sequence of speech syllable codes of which a language word is composed such that these syllable based language words are sorted in the order of their syllables from speech pronunciation; (c) the second field storing equivalent language words such that the database can be searched, based on the syllable based language word, and its equivalent language word found.

Claim 23

The speech syllable based dictionary as in claim 19, wherein, the fields further comprising: an additional field storing a representation of grammatical attribute, within the sentence structure, of the language word.