EP0886853B1 - Microsegment-based speech-synthesis process - Google Patents

Abstract

A digital speech synthesis process in which utterances in a language are recorded, and the recorded utterances are divided into speech segments which are stored so as to allow their allocation to specific phonemes. A text which is to be output as speech is converted to a phoneme chain and the stored segments are output in a sequence defined by the phoneme chain. An analysis of the text to be output as speech is carried out and thus provides information which completes the phoneme chain and modifies the timing sequence signal for the speech segments which are to be strung together for output as speech. The process uses microsegments consisting of: segments for vowel halves and semi-vowels and extending as far as the vowel middle, and a second vowel half from the vowel middle to just before the vowel end; segments for quasi-stationary vowel components cut from the middle of a vowel; consonant segments beginning shortly before the front phoneme boundary and ending shortly before the rear phoneme boundary; and segments for vowel-vowel sequences cut from the middle of a vowel-vowel transition.

Description

The invention relates to a digital speech synthesis method according to the preamble of claim 1.

In the synthetic generation of speech with computers essentially three methods are known.

In the formant synthesis with a Excitation source with downstream filters Resonance properties of the human extension tube and their changes in speaking caused by the Movements of the articulation organs are caused replicated. These resonances are characteristic for the structure and perception of vowels. For The first three limit the computing effort up to five formants of a speech sound synthetically with the excitation source. In this type of synthesis therefore only for the different excitation waveforms a small space requirement in a computer to provide. Furthermore, a simple change from Duration and fundamental frequency excitation waveforms realized become. The disadvantage, however, is that for voice output an extensive control apparatus is needed, which often the use of digital processing processors makes necessary. Another disadvantage is that the output speech sounds unnatural and metallic and particular weaknesses in nasal and Obstruents, d. H. Plosives / p, t, k, b, d, g /, Affrikaten / pf, ts, tS / und Frikativen / f, v, s, z, S, Z, C, j, x, h /.

In this text, they put between slashes // Letters represented by symbols represent SAMPA notation, see: Wells, J .; Barry, W.J .; Grice, M .; Fourcin, A .; Gibbon D. (1992); Standard computer compatible Transcription, in: ESPRIT PROJECT 2589 (SAM) Multi-lingual speech input / output assessment, methodology and standardization; Final report; Doc. SAM-UCL-037, pages 29ff.

In articulatory synthesis, the modeled acoustic conditions in the extension tube, see above that the articulatory positions and movements be simulated mathematically when speaking. It will So an acoustic model of the extension pipe is calculated, which leads to a considerable computational effort and a requires large computing capacity. Still it sounds like that automatically generated language unnatural and technical.

In addition, the concatenation synthesis is known with parts of real spoken utterances like that be chained that new expressions arise. The So individual parts of the language form building blocks for Speech generation. The size of the parts can - depending by application - from words and phrases to Excerpts from sounds are sufficient. For the artificial Generate language with unlimited vocabulary are offered as units of half syllables or smaller ones Cutouts. Larger units only make sense when a limited vocabulary is synthesized should.

In systems that do not require resynthesis the choice of the correct cutting point of the Language modules crucial for the quality of the Synthesis. The key here is melodic and spectral To avoid breaks. Concatenative synthetic processes then achieve - especially with large building blocks - a more natural sound than the other methods. The standard effort for the generation of the sounds is also quite low. The limitations of this Procedures are in the relatively large space requirement for the required language modules. Another The limitation of this procedure is that once recorded modules in the known systems only with complex resynthesis processes (e.g. in duration or frequency) that can be changed detrimental to the speech sound and intelligibility impact. There are therefore several different ones Variants of a language module added what increases storage space requirements.

Among the concatenation synthesis processes are in essential four synthetic methods known to it allow language without restricting vocabulary to synthesize.

Phonesis involves the concatenation of sounds or Phones made. For Western European languages with a sound inventory of approx. 30-50 sounds and one average duration of the sounds is approx. 150 ms the storage space requirement is manageably small. Indeed these speech signal modules are missing perceptually important transitions between the individual sounds that even incomplete by blending out individual ones Loud or more complex resynthesis processes can be modeled. Hence this Type of synthesis qualitatively unsatisfactory. Also the Consideration of the phonetic context of individuals Lute by dropping sound variants of a Loud in their own speech signal modules in the so-called allophone synthesis improves this Speech result due to non-compliance with the articulatory-acoustic dynamics not essential.

The most common form of concatenation synthesis is Diphone synthesis; this uses signal modules from the middle of an acoustically defined speech up to range to the middle of the next speech. Thereby the perceptually important transitions from one According to the other, considered as acoustic Follow the movements of the speech organs in the speech signal occur. It also becomes the signal building blocks at spectrally relatively constant locations put together what the potentially existing Disturbances in the signal flow at the joints of the individual Diphone decreased. The sound inventory of western European Languages consist of 35 to 50 sounds. For one language with 40 sounds, there are theoretically 1600 Diphone pairs, which are then identified by phonotactic Restrictions can actually be reduced to around 1000. In natural language differ unstressed and emphasized sounds both in sound and in duration from each other. To these differences in synthesis Adequate to be considered in some systems for stressed and unstressed sound sequences different Diphone added. Depending on the approach, there will be 1000 to 2000 diphones with an average duration of approx. 150 ms is required, depending on the Dynamic and signal bandwidth requirements Storage space required for the signal blocks of up to 23 MB results. A typical value is around 8 MB.

On a similar principle as the diphone synthesis are also based on triphone and half syllable synthesis. Here, too, the cutting point is in the middle of the Lute. However, larger units are recorded which takes larger phonetic contexts into account can be. The number of combinations increases however proportional to. In half syllable synthesis is an intersection for the units used in the middle of the vowel of a syllable. The other cutting point is at the beginning or end of a syllable, so depending on the structure of the syllable also sequences of several Consonants can be included in a language module. In German there are about 52 different sound sequences in initial syllables of morphemes and about 120 sound sequences counted for medial or final syllables of morphemes. This results in a theoretical number of 6240 Half syllables for German, some of which are uncommon. Since half-syllables are usually longer as a diphone, the storage space requirement for the Speech signal modules that the diphones a lot.

The biggest problem is therefore with a qualitative high quality speech synthesis system of considerable Space requirements. To reduce this need For example, it was suggested that the silence in Closure of plosives for all plosives use. From EP 0 144 731 B1 is a Speech synthesis system known in which parts of diphones can be used for several sounds. There will be a Described speech synthesizer, the unit speech waveforms, created by sharing a double sound be saved and certain expression symbols equates. A synthesizer reads the Unit speech waveforms corresponding to the Output symbols of the converted sequence from Expression symbols from memory. Based on the The speech part of the input characters determines whether two read unit speech waveforms either directly be connected when the input speech part of the Input character is unvoiced, or a given one first interpolation method is used if the input speech part of the input times voiced is the same unit waveform for both a voiced / g, d, b / as well as for his corresponding voiceless / k, t, p / sound used becomes. Furthermore, unit speech waveforms should also be in the memory that the one Consonants following the vowel part or the one Represent consonants preceding vowel part. The Transitional areas from a consonant to a vowel or from a vowel to a consonant can in each case for the consonants k and g, t and d as well as p and b be equated. The space requirement will be thus reduced, but requires the specified Interpolation process a not insignificant Computing effort.

DE 27 40 520 A1 describes a method for synthesis known by language, in which each phoneme of in one Memory stored phoneme elements is formed taking periods of sound vibrations from natural Language won or artificially synthesized. Of the Text to be synthesized becomes grammatical sentence by sentence and analyzed phonetically according to the rules of language. In addition to the periods of sound vibrations are everyone Phoneme specific types and a number of time periods of noise phonemes with corresponding Duration, amplitudes and spectral distribution are compared. The periods of sound vibrations and the Elements of the noise phonemes are in digital form as Sequence of amplitude values of the corresponding Vibration is stored in a memory and are stored in the Reading process according to the frequency characteristic and changed to achieve the naturalness of language.

This results in a digital speech synthesis process according to the concatenation principle according to the The preamble of claim 1 is known.

To use as little memory as possible get along, according to the synthetic method of DE 27 40 520 A1 individual periods of sound vibrations stored with characteristic formant distribution. That every phoneme while keeping the basic characteristics of the set certain types and numbers of the stored Periods of sound vibrations are determined and then together form the acoustic speech impression. After that, extremely short time series elements of the length of a period of the fundamental wave a sound from the memory and depending on the previous number of reproduced consecutively repeated. For realizing smooth phoneme transitions periods (synthetic) with formant distributions, the transition between the phonemes correspond, used or the amplitudes in the range of the transitions concerned.

The disadvantage is that a sufficient naturalness of the Voice playback due to multiple playback same period pieces, possibly only shortened synthetically or extended, is not reached. Furthermore, the significantly reduced memory requirements by one bought more analysis and interpolation effort, what computing time costs.

A for the speech synthesis method of DE 27 40 520 A1 A similar process is known from WO 85/04747, in the case of a completely synthetic Generation of the language segments is assumed. The Segments of speech that represent phonemes or transitions are made from synthetic waveforms following a predetermined manner several times, possibly in length abbreviated and / or reproduced in voices, generated. Especially with the phoneme transitions also from an inverted reproduction of certain Made use of time series. Another disadvantage is that with significantly reduced space requirements due to extensive analysis and Synthesizing processes a considerable computing capacity is needed. The speech reproduction is nevertheless missing the natural variance.

The object of the invention is therefore based on the DE 27 40 520 A1 to specify a speech synthesis method, in the case of small storage space requirements without high Computing effort a high quality speech output is achieved.

This task is solved with a speech synthesis process according to claim 1.

1.

Segmente für Vokalhälften und Halbvokalhälften: Sie geben in der Dynamik der spektralen Struktur die Bewegungen der Sprechorgane von bzw. zu der Artikulationsstelle des benachbarten Konsonanten an. Aufgrund der Silbenstruktur der meisten Sprachen ist häufig eine Konsonant-Vokal-Konsonant-Folge anzutreffen. Da die Bewegungen der Sprechorgane für eine gegebene Artikulationsstelle entsprechend den relativ unbeweglichen Teilen des menschlichen Ansatzrohres unabhängig von der Artikulationsart, d. h., unabhängig von den vorangehenden oder nachfolgenden Konsonanten, vergleichbar sind, ist daher für jeden Vokal nur ein Mikrosegment pro globaler Artikulationsstelle des vorherigen Konsonanten (= erste Hälfte des Vokals) und ein Mikrosegment pro Artikulationsstelle des folgenden Konsonanten (= zweite Hälfte des Vokals) nötig.

2.

Segmente für quasi stationäre Vokalteile: Diese Segmente sind aus der Mitte von langen Vokalrealisierungen, die klanglich relativ konstant wahrgenommen werden, herausgetrennt. Sie werden in verschiedenen Textpositionen bzw. Kontexten eingesetzt, beispielsweise am Wortanfang, nach den Halbvokalsegmenten, die bestimmten Konsonanten bzw. Konsonantfolgen folgen, im Deutschen beispielsweise nach /h/, /j/ sowie /?/, zur Enddehnung, zwischen nicht diphthongischen Vokal-Vokalfolgen und in Diphthongen als Start- und Zielpositionen.

3.

Konsonantische Segmente: Die konsonantischen Segmente sind so gebildet, daß sie unabhängig von der Art der Nachbarlaute für mehrere Vorkommen des Lautes entweder generell oder wie vornehmlich bei Plosiven im Kontext von bestimmten Lautgruppen verwendet werden können.

With the speech synthesis method according to the invention, a generalization is achieved when using the speech signal modules in the form of microsegments. The use of a separate acoustic segment for each of the possible connections of two speech sounds, which is necessary in diphone synthesis, is thus avoided. The microsegments required for voice output can be broken down into three categories. These are:

1.

Segments for vowel halves and half vowel halves: They indicate the movements of the speech organs from or to the articulation point of the neighboring consonant in the dynamics of the spectral structure. Due to the syllable structure of most languages, a consonant-vowel-consonant sequence can often be found. Since the movements of the speech organs for a given articulation point are comparable according to the relatively immovable parts of the human extension tube regardless of the articulation type, i.e. independent of the preceding or following consonants, there is therefore only one microsegment for each vowel per global articulation point of the previous consonant ( = first half of the vowel) and one microsegment per articulation point of the following consonant (= second half of the vowel) required.

2nd

Segments for quasi-stationary vowel parts: These segments are separated from the middle of long vowel realizations, which are perceived relatively constant in terms of sound. They are used in various text positions or contexts, for example at the beginning of a word, after the semi-vowel segments that follow certain consonants or consonant sequences, in German for example after / h /, / j / and /? /, For the final stretching between non-diphthongic vowels. Vowel sequences and in diphthongs as start and target positions.

3rd

Consonant segments: The consonant segments are formed in such a way that, regardless of the type of neighboring sounds, they can be used for several occurrences of the sound either generally or, as is the case with plosives, in the context of certain sound groups.

It is important that the divided into three categories Microsegments multiple times in different phonetic Contexts can be used. That is, at Sound transitions the perceptually important transitions from one sound to another without being considered that for each of the possible connections of two Speech loud own acoustic segments required are. The division according to the invention into microsegments, that shares a sound transition enables use identical segments for different sound transitions for a group of consonants. With this principle of Generalization when using Speech signal modules is the one for storing the Speech signal modules reduced memory space required. Still, the quality is synthetic output language due to the consideration of the perceptually important sound transitions very well.

The fact that the segments for vowel halves and Half vowel halves in a consonant-vowel or vowel-consonant sequence for each of the articulation points of the neighboring consonants, namely labial, alveolar or velar, are the same for the language segments for Vowels a multiple use of the micro segments for allows different phonetic context and thus a considerable reduction in storage space reached.

If the segments for quasi stationary vowel parts are intended for vowels at the beginning of words and Vowel-vowel sequences, with a small number of additional microsegments a considerable Sound improvement of the synthetic language for Word beginnings, diphthongs or vowel-vowel sequences reached.

The fact that the consonant segments for Plosive are divided into two microsegments, a first segment, which includes the closing phase and a second segment, that includes the solution phase becomes a further generalization of the language segments reached. Especially the closure phase can be carried out for all plosives represent a time series of zeros. For this part the sound reproduction is therefore no storage space required.

The solution phase of the plosive is in context differentiated the following sound. It can be another Generalization can be achieved in solving for vowels only after the following four vowel groups - front, unrounded vowels; front, rounded Vowels; deep or centralized vowels and back, rounded vowels - and with a solution to consonants only after three different articulation points, labial, alveolar or velar, is distinguished so that for example, 42 microsegments for the German language for the six plosives / p, t, k, b, d, g / zu three consonant groups by articulation point and to four vowel groups must be saved. This reduced due to the multiple use of the Micro segments for different phonetic context the storage space requirement further.

It is advantageous to shorten vowel segments a vowel segment from an articulation point runs to the middle of the vowel, the start and at one Vowel segment that goes from the middle of the vowel to the following articulation point that Target position always reached while moving to or shortened from the "vowel center". Such Shortening the microsegment forms, for example unstressed syllables after, in the natural, deviations from the flowing speech spectral target quality of the respective vowel are reproduced and thus the naturalness of the Synthesis is increased. It is also advantageous that for such linguistic variations already saved segments none corresponding to the segment further storage space is required.

With the analysis of the text to be output as language will manipulate the microsegments in Dependency of the analysis result reached. In order to can vary the pronunciation depending on the Sentence structure and semantics both sentence by sentence and be reproduced word for word in the sentences without that additional microsegments for different Debates are necessary. The storage space requirement can thus be kept low. It also requires the manipulation in the time domain is not expensive Arithmetic operations. Nevertheless, with the Speech synthesis techniques produced a very language natural character.

In particular, using the analysis on the as language Text to be output, pauses in speech are recognized. The Phoneme chain is closed at these points with pause symbols a symbol chain added, whereby in the sequence of the micro segments on the digital break symbols Zeros are inserted in the time series signal. The additional information about a break point and their pause duration is due to the sentence structure and predetermined rules determined. The pause will be by the number of digital zeros to be inserted in Dependency of the sampling rate realized.

By recognizing phrase boundaries with the analysis and the phoneme chain at these points Expansion symbols are added to form a symbol chain, where the sequence of the microsegments Micro segments corresponding to the symbols one Experiencing the length of play in the time domain can be Final stretching in the synthetic Speech reproduction are reproduced. This Manipulation in the time domain is already at the assigned microsegments executed. It will therefore no additional language modules to implement Final stretches required what the space requirements keeps low.

Because the analysis recognizes stresses and the phoneme chain at these points with stress symbols for different stress values to one Symbol chain is added, being in the sequence of the microsegments on the microsegments with Stress symbols change the duration of the Speech sounds are made in natural language reproduced existing types of stress. The Main information regarding the by the playing time word accent formed is in a lexicon. The then for sentence accents to be selected with intonation Emphasis is placed on the analysis of language Text to be output from the sentence structure and predetermined rules determined. Depending on the determined Emphasis is placed on the relevant microsegment or by omitting certain microsegment sections abbreviated. To generate a versatile language with justifiable Computing effort has five reduction levels for vocal microsegments proved to be sufficient, so that a total of six playing time options To be available. These cut levels are on the previously saved microsegment and are marked depending on the context of the text analysis according to the Analysis result, d. H. of the one to be chosen Emphasis, driven.

Both the duration of the play in phrase finals Syllables, as well as the various reduction levels for Emphasis can prefer the same Reduction levels can be realized in the microsegments. In contrast to stressed syllables, in which the temporal expansion evenly on all microsegments is distributed in the end syllables of phrases, namely of language units, for example in the Written language with the punctuation marks comma, semicolon, Dot and colon are noted, a progressive Extension of the playing time planned. this will achieved by increasing the playing time of the Microsegments in the phrase final syllables from second microsegment by one level each.

For example, the sentence "He lived in Paris." the last syllable "-lives", pronounced / vo: nt /, stretched so that the microsegment chain shown in the table in the first line with the normal continuous step indicated in brackets, if this syllable is not at the end of the phrase, in accordance with the stretch symbols microsegment chain shown in the third line is transferred. The range of values for the expansion levels goes from 1-6, whereby larger numbers correspond to a longer duration. The% symbol does not create a permanent change. normal [2v] o v [5o] [5o] n [2n] t t [2t] [2t]. . . symbol % % +1 +2 +3 +4 stretched [2v] o v [5o] [6o] n [4n] t t [5t] [6t] ...

Education in other languages or dialects is similar. In English, for example, the final stretch would be from the sentence "He saw a shrimp." for the last word are formed by microsegments as follows: normal [2S] r [2r] I r [3I] [3I] m [2m] p p [2p] [2p] ... symbol % % % +1 +2 +3 +4 stretched [2S] r [2r] I r [3I] [4I] m [4m] p p [5p] [6p] ...

In the case of open syllables, ie those ending with a vowel, such as "He was there.", The playing time of the second microsegment is pronounced from "there" / there: /, by two steps. normal d [2d] [2d] a d [4a] [4a] ... symbol % % % +2 stretched d [2d] [2d] a d [4a] [6a] ...

This procedure is carried out until the longest continuous level (= 6) is reached.

By assigning intonations to the analysis and the phoneme chain at these points Intonation symbols are added to form a symbol chain, being in the sequence of the microsegments a change in fundamental frequency in the intonation symbols certain parts of the periods of microsegments in Time range is performed, the melody reproduced linguistic utterances. The The fundamental frequency change is preferably carried out by skipping and adding certain Samples. For this, the previously recorded voiced microsegments, i.e. Vowels and sonorants, marked. Every voting period is automatically included the spectrally important first part, in which the vocal folds are closed, and the less important second part, in which the vocal folds are open, treated separately. The markings are set so that only the spectrally noncritical in the signal output second part of each period Fundamental frequency change shortened or extended are reproduced. This will take up space for the reproduction of intonations at the Not significantly increased and the Computational effort due to the manipulation in the time domain kept low.

When chaining different microsegments together speech synthesis is largely free of interference acoustic transition between successive Micro segments achieved in that the Microsegments with the first sample after the first positive zero crossing, d. H. a zero crossing with positive signal rise, start and start with the last one Sample before the last positive zero crossing end up. The digitally stored time series of Micro segments are thus almost continuously in line to each other. So are due to digital leaps resulting cracking noises avoided. You can also represented at any time by digital zeros Closing phases of plosives or word breaks and general language breaks essentially steady be inserted.

The following is an embodiment of the invention described in detail with reference to the drawings.

Fig. 1

ein Ablaufdiagramm des Sprachsyntheseverfahrens,

Fig. 2

ein Spektrogramm und Zeitsignal des Wortes "Phonetik" und

Fig. 3

das Wort "Frauenheld" im Zeitbereich.

It shows:

Fig. 1

a flow diagram of the speech synthesis process,

Fig. 2

a spectrogram and time signal of the word "phonetics" and

Fig. 3

the word "womanizer" in the time domain.

The process steps of the speech synthesis system according to the invention are in a flowchart in Fig. 1 shown. The input for the speech synthesis system is a text, for example a text file. The Words in the text are saved using a computer stored lexicons assigned a phoneme chain that represents the pronunciation of each word. In the language, especially in the German language, word formation often takes place through composition of words and parts of words, e.g. with pre and Suffixes. The pronunciation of words like "house building", "Development", "buildable" etc., can be from one trunk, here "bau", derived and with the pronunciation of the Vorund Suffixes are connected. You can also Connection sounds such as "s" in "bailiff", "es" in "State sports school" and "n" in "miners", be taken into account. Thus, in the event that one word is not in the lexicon, different Replacement mechanisms to correct the pronunciation of the word to verify. It tries to do that first searched word from partial entries of the lexicon, as above described, put together. If not succeeds, an attempt is made via a syllable dictionary in which Syllables with their pronunciations are registered to one Pronunciation. If this also fails, there is Rules, like sequences of letters in phoneme sequences are to be implemented.

Under the phoneme chain created as shown above 1 is the syntactic-semantic analysis shown. There are in addition to the known ones Pronunciation information in the lexicon syntactic and contain morphological information together local with certain keywords of the text linguistic analysis allow the phrase boundaries and output accented words. Based on these Analysis will be the phoneme chain resulting from the pronunciation information of the lexicon comes, modified and additional information about break times and Pitch values of the microsegments are inserted. It a phoneme-based, prosodically differentiated Symbol string, which is the input for the actual Delivers speech.

• der phonologischen Länge eines Lautes, die bei jedem Phonem bezeichnet ist, beispielsweise /e:/ für langes 'e' in /fo'ne:tIK/,
• der Akzentuierung der Silbe, die in der Phonemkette vor der betonten Silbe bezeichnet ist, beispielsweise, /fo'ne:tIK/,
• den Regeln für phrasenfinale Dehnung und
• ggf. andere Regeln, die auf der Abfolge von akzentuierten Silben beruhen, wie beispielsweise die Längung von zwei betonten aufeinanderfolgenden Silben.

For example, the syntactic-semantic analysis takes into account word accents, phrase boundaries and intonation. The gradations of the emphasis of syllables within a word are marked in the lexicon entries. The emphasis levels are thus specified for the reproduction of the microsegments forming this word. The stress level of the microsegment of a syllable results from:

• the phonological length of a sound, which is designated for each phoneme, for example / e: / for long 'e' in / fo'ne: tIK /,
• the accentuation of the syllable, which is indicated in the phoneme chain before the stressed syllable, for example, / fo'ne: tIK /,
• the rules for phrase final stretching and
• if necessary, other rules based on the sequence of accented syllables, such as the elongation of two stressed syllables in succession.

The phrase boundaries at which in addition to certain intonational courses the end of the phrase takes place through linguistic analysis determined. The sequence of parts of speech becomes with given rules determines the limit of phrases. The implementation of the intonation is based on a Intonation and pause description system in which basically between intonation courses that Phrase boundaries take place (rising, falling, constant, falling-rising) and those that around Accents are localized (low, high, rising, falling). The assignment of the Intonation courses are based on the syntactic and morphological analysis under Inclusion of certain keywords and characters in the text. For example, have questions with Bursting position (recognizable by the question mark on End and information that the first word of the sentence a finite verb is) a deep accent and one high rising limit tone. Normal statements have one high accent tone and a falling final phrase limit. The course of the intonation is according to given Rules generated.

For the actual speech output, the phoneme-based symbol chain in a microsegment sequence converted. The conversion of a sequence of two Phonemes in microsegment sequences are made using a Rule set in which each phoneme sequence is a sequence of Micro segments is assigned.

It is in the sequence of the by Microsegment chain specified consecutive Microsegments the additional information about Emphasis, pause duration, final stretch and intonation considered. The modification of the microsegment sequence takes place exclusively in the time domain. In the time series signal of the lined up micro segments is, for example, a language break Insert digital zeros at the by one corresponding pause symbol marked position realized.

The voice output then takes place through digital / analog conversion the manipulated time series signal, for example, one arranged in the computer "Soundblaster" card.

Fig. 2 shows a spectrogram in the upper part and lower part the associated time signal for the Word example "Phonetics". The word "phonetics" is used in Symbols as a phoneme sequence between slashes like shown as follows / fone: tIk /. This phoneme sequence is on the abscissa representing the time axis in the upper Part of Fig. 2 applied. The ordinate of the Spectrogram of Fig. 2 denotes the frequency content of the speech signal, the degree of blackening to Amplitude of the corresponding frequency is proportional. Corresponds to the time signal shown above in FIG. 2 the ordinate of the current amplitude of the signal. in the middle fields are the with vertical lines Microsegment boundaries shown. The specified therein Abbreviations give the designation or Symbolization of the respective microsegment. The Example word "phonetics" thus consists of twelve Microsegments.

The names of the microsegments are chosen so that the lute outside the parentheses the context mark, in the brackets the sounding sound is specified. It will be the contextual Transitions of the speech sounds are taken into account.

The consonant segments ... (f) and (n) e are on segmented the respective sound limit. The Plosive / t / and / k / are in a locking phase (t (t) and k (k)), the digitally by sampling values set to zero is reproduced and used for all plosives, and a short solution phase (here: (t) I and (k) ...), which is context sensitive, split. The vowels are each divided into vowel halves, with the intersection points at the beginning and in the middle of the vowel.

In Fig. 3 is another word example "womanizer" in Played time range. The phoneme sequence is with / fraU @ nhElt / specified. The word shown in Fig. 3 comprises 15 microsegments, here also quasi stationary microsegments occur. The first two Microsegments ... (f) and (r) a are consonant Segments whose context is only on one side is specified. After the semi-vowel r (a), the one Transition of the velar articulation point to the middle of the a includes, to form the diphthong / aU / die Starting position a (a). AU (AU) includes the perceptual important transition between the start and the Target position u (U). (U) @ contains the transition from / U / after / @ /, which is usually followed by @ (@) ought to. This would make / @ / take too long, so this The segment for / @ / and / 6 / is omitted for permanent reasons and only the second vowel half (@) n is played. (n) h represents a consonant segment. The transition from consonants to / h / - unlike vowels - not specified. Therefore there is no segment n (h). (h) E contains the breathed part of the vowel / E /, the followed by the quasi-stationary E (E). (E) 1 contains the second vowel half of / E / with the transition to the dental articulation site. E (1) is a consonant microsegment, in which only the Precontext is specified. The / t / is divided into a closure phase t (t) and a solution phase (t) ..., that goes to silence (...).

According to the invention, the large number of possible articulation points is limited to three essential areas. The grouping is based on the similar movements carried out by the articulators to form the sounds. Because of the comparable articulator movements, the spectral transitions between the sounds are similar within the three groups listed in Table 1. Articulators and articulation points and their names Summary description Articulator Articulation point labial bilabial bottom lip Upper lip labiodental bottom lip upper incisors alveolar dental alveolar Tongue tip upper incisors Tip of the tongue or tongue Tooth embankment, alveoli velar palatal front of the tongue hard palate, palatum velar medium back of the tongue soft palate, velum uvular back of the tongue Suppository, uvulum - pharyngeal Tongue root back pharynx wall glottal Vocal fold Vocal fold

jeweils dieselben zwei Vokalhälften verwendet werden, weil der Anfangskonsonant jeweils mit dem Verschluß der beiden Lippen (bilabial) und der Endkonsonant durch Anhebung der Zungenspitze zum Zahndamm (= alveolar) gebildet werden. Neben der labialen und der alveolaren gibt es noch die velare Artikulationsstelle. Eine weitere Generalisierung wird durch die Gruppierung der postalveolaren Konsonanten /S/ ( wie in Ma sch e) und /Z/ (wie in Ga g e) zu den alveolaren und der labiodentalen Konsonaten /f/ und /v/ mit den labialen erreicht, so daß, wie oben angegeben, auch /fa(tS)/, /va(tS)/, /fa(dZ)/ und /va(dZ)/ dieselben Vokalsegmente enthalten können. Für die Mikrosegmente der o.g. Beispielsilben gilt also: p(a) = b(a) = m(a)a = (pf)(a) = f(a) = v(a) und
(a)t = (a)d = (a)s = (a)z = (a)(ts) = (a)(tS) = (a)(dZ)
= (a)n = (a)1.Therefore, only one microsegment per articulation point of the previous consonant (= 1st half of the vowel) and one microsegment per articulation point of the following consonant (= 2nd half of the vowel) is used for each vowel. It can e.g. B., for the syllables

the same two vowel halves are used because the initial consonant is formed with the closure of the two lips (bilabial) and the final consonant is formed by raising the tip of the tongue to the perineum (= alveolar). In addition to the labial and alveolar, there is also the velar articulation point. A further generalization is achieved by grouping the postalveolar consonants / S / (as in Ma sch e) and / Z / (as in Ga g e) to the alveolar and labiodental consonates / f / and / v / with the labial, so that, as stated above, / fa (tS) /, / va (tS) /, / fa (dZ) / and / va (dZ) / can also contain the same vowel segments. For the microsegments of the above-mentioned example syllables, the following applies: p (a) = b (a) = m (a) a = (pf) (a) = f (a) = v (a) and
(a) t = (a) d = (a) s = (a) z = (a) (ts) = (a) (tS) = (a) (dZ)
= (a) n = (a) 1.

• die ersten Hälften der Monophthonge
  /i:, I, e:, E, E:, a(:), O, o:, U, u:, y:, Y, 2:, 9, @, 6/, die nach einem labial, alveolar bzw. velar gebildeten Laut auftreten;
• die zweiten Hälften der Monophthonge
  /I:, I, e:, E, E:, a(:), O, o:, U, u:, y:, Y, 2:, 9, @, 6/ vor einem labialen, alveolaren oder velaren Laut;
• Erste und zweite Hälften der Konsonanten /h/ und /j/ aus den Kontexten:
  • nicht-offener ungerundeter Vordervokal /i:, I, e, E, E:/,
  • nicht-offener gerunder Vordervorkal /y:, Y, 2:, 9/,
  • offener ungerundeter zentrale Vokal /a(:), @; 6/,
  • nicht-offener gerunderter Hinterzungenvokal /O, o:, U, u:/.

In addition to the vowel halves for vowel "a" just described, the following microsegments also belong to the category of vowel halves and half vowel halves:

• the first halves of the monophthongs
  / i :, I, e :, E, E :, a (:), O, o :, U, u :, y :, Y, 2 :, 9, @, 6 /, which after a labial, alveolar or velar-formed sound occur;
• the second halves of the monophthongs
  / I :, I, e :, E, E :, a (:), O, o :, U, u :, y :, Y, 2 :, 9, @, 6 / in front of a labial, alveolar or velar Loud;
• First and second halves of the consonants / h / and / j / from the contexts:
  • non-open, unrounded front vowel / i :, I, e, E, E: /,
  • non-open round front leading / y :, Y, 2 :, 9 /,
  • open unrounded central vowel / a (:), @; 6 /,
  • non-open rounded tongue vowel / O, o :, U, u: /.

• wortinitial,
• nach den Halbvokalsegmenten /h/, /j/ sowie um /?/,
• zur Enddehnung, wenn auf einer Endsilbe komplexe Tonbewegungen realisiert werden müssen,
• zwischen nicht diphthongischen Vokal-Vokal-Folgen, sowie
• in Diphthongen als Start- und Zielpositionen.

In addition, segments for quasi-stationary vowel parts are required to simulate the middle of a long vowel realization. These microsegments are used in the following positions:

• word initial,
• according to the semi-vowel segments / h /, / j / and by /? /,
• for final stretching when complex sound movements have to be realized on a final syllable,
• between non-diphthongic vowel-vowel sequences, as well
• in diphthongs as start and target positions.

The multiple use of the micro segments in different phonetic contexts is used in the Diphone synthesis resulting multiplication effect of Sound combinatorics considerably reduced without the Impair the dynamics of articulation.

In the generalization shown according to the invention in the language modules it is theoretically possible for the German language with a number of 266 microsegments get along, namely 16 vowels at 3 articulation points, stationary, over; 6 plosives to 3 Consonate groups by articulation point and 4 Vowel groups; / h /, / j / and /? / to more differentiated Vowel groups. To improve the sound quality of the synthetically formed language should be the number of required microsegments for the German language each after sound differentiation are between 320 and 350. This corresponds to the relatively short time Micro segments with a storage space requirement of approx. 700 kB at 8 bit resolution and 22 kHz sampling rate. That delivers a reduction compared to the known diphone synthesis by a factor of 12 to 32.

To further improve the sound of the synthetic educated language is intended in the individual To attach microsegment markings that a Shortening, stretching or frequency change on the microsegment allow in the time domain. The markings are at the zero crossings with a positive slope of the time signal of the microsegments. All in all five reduction levels are carried out, so that Microsegment together with the unabridged rendering has six different levels of play time. Both The cuts are made so that for a vowel segment, that from an articulation point to the middle the vowel starts, and at one Vowel segment that goes from the middle of the vowel to the following articulation point that Target position (= articulation point of the following Consonants) is always achieved during the movement to or from the "vocal center" is shortened. By this procedure becomes another generalized one Allows use of the microsegments. The same Signal modules provide the basic elements for long and short sounds in both stressed and unstressed Syllables. The reductions in the rate not accented words are also used by the same in sentence-accentuated position recorded microsegments derived.

In addition, the intonation can be linguistic Comments by a change in the fundamental frequency of the periodic parts of vowels and sonorants become. This is done through a fundamental frequency manipulation performed in the time domain on the microsegment, although hardly tonal losses arise. The spectral Information-important part (1st part = phase of the closed glottis) of each voting period and the less important second part (= phase of the open glottis) are treated separately. The first voting period and the one to be kept constant "closed phase" (1st part of the period) marked. Leave because of the monotonous way of speaking all other periods in the microsegment automatically find and thus define the closed phases. When the signal is output, the spectrally non-critical ones "Open phases" proportional to the frequency increase spent shorter, which is a shortening of the Total periods. When the frequency is reduced, the open phase extended in proportion to the degree of subsidence. Frequency increase and decrease are over a Microsegment performed uniformly. The thereby in Gradual intonation is due to the natural "Auditory integration" of the hearing person largely smoothed. In principle it is possible, however Frequencies also within a microsegment change, up to the manipulation of individual periods.

Below is the inclusion and segmentation of Micro segments as well as the speech reproduction are described.

Individual words that represent the corresponding sound combinations are monotonous and emphasized by one person spoken. These real spoken utterances will be recorded and digitized. From these digitized The micro-segments become utterances cut out. The intersection of the consonant segments are chosen so that the Influence of neighboring sounds at the microsegment borders is minimized and the transition to the next sound is not is more precisely perceptible. The vowel halves are made out cut from the surroundings by voiced plosives, taking noisy parts of the locking solution be eliminated. The quasi-stationary vowel parts are separated from the middle by long sounds.

All segments are thus from the digital signal of the they contain the statement that they cut with the first sample after the first positive Start zero crossing and start with the last sample end before the last positive zero crossing. In order to crackling noises are avoided.

The digital signal has to limit the memory requirement for example a bandwidth of 8 bits and a sampling rate of 22 kHz.

The microsegments separated in this way are addressed according to the sound and context and stored in a memory.

A text to be output as language is accompanied by the corresponding Address order fed to the system. The order of sounds determines the selection of Addresses. According to this order of addresses the microsegments are read from the memory and strung together. This digital time series is in a digital / analog converter, for example in one so-called Soundblaster card, in an analog signal converted that through speech devices, for example a speaker or headphones, can be spent.

The speech synthesis system according to the invention can be based on an ordinary PC, where a RAM of about 4 MB is sufficient. The one with the System realizable vocabulary is practical unlimited. The language is easy to understand, the computational effort for modifications of the Microsegments, for example cuts or Fundamental frequency changes, is small since that Voice signal is processed in the time domain.

Claims (15)

1. A digital speech-synthesis process , in which utterances of a language are recorded first, the recorded utterances are divided in speech segments, and the segments are stored allocated to defined phonemes, a text to be output as speech then being converted into a phoneme string and the stored segments are successively output in a sequence defined by said phoneme string, and an analysis of the text to be output as speech is carried out and thus provided information supplementing the phoneme string, such information modifying the series of statistical values signal of the speech segments to be concatenated for the speech output, characterized in that microsegments are used as speech segments, such microsegments consisting of:

   Segments for vowel halves and semi-vowel halves, vowels between consonants being split into two microsegments, a first vowel half beginning shortly after the beginning of the vowel and extending up to the middle of the vowel, and a second vowel half extending from the middle of the vowel up to just before the end of the vowel;

   Segments for quasi-stationary vowel parts, such segments being cut from the middle of a vowel;

   Consonantal segments beginning shortly after the front sound boundary and ending shortly before the rear sound boundary; and

   Segments for vowel-vowel sequences, which are cut from the middle of a vowel-vowel transition.
2. The speech-synthesis process according to claim 1, characterized in that the segments for vowel halves and semi-vowel halves in a consonant-vowel or vowel-consonant sequence are identical for each of the articulation places of the adjacent consonant, namely labial, alveolar or velar.
3. The speech-synthesis process according to claim 1 or 2, characterized in that the segments for quasi-stationary vowel parts are provided for vowels at the beginnings of words and for vowel-vowel sequences as well as for the sounds /h/, /j/ and glottal stops.
4. The speech-synthesis process according to claim 1, 2 or 3, characterized in that the consonantal segments for plosives are divided in two microsegments: a first segment comprising the closure phase, and a second segment comprising the release phase.
5. The speech-synthesis process according to claim 4, characterized in that the closure phase is reached for all plosives by lining up digital zeros.
6. The speech-synthesis process according to claim 4 or 5, characterized in that the release phases of the plosives are differentiated after the sound following in the context as follows:
   Release into vowels:

   Front, unrounded vowels;

   front, rounded vowels;

   low or centralized vowels; and

   back, rounded vowels; as well as

   release into consonants according to the global articulation place:

   labial;

   alveolar; and

   velar.
7. The speech-synthesis process according to claim 1, 2, 3, 4, 5 or 6, characterized in that the analysis detects speech pauses and the phoneme string is extended to a symbol string by adding pause symbols at speech pauses, digital zeros being inserted in the series of statistical values signal on the pause symbols when the microsegments are concatenated.
8. The speech-synthesis process according to claim 1, 2, 3, 4, 5, 6, or 7, characterized in that the analysis detects phrase boundaries and that the phoneme string is extended in said places with lengthening symbols to form a symbol string, a lengthening of the playback duration taking place on the markings within the time domain when the microsegments are concatenated.
9. The speech-synthesis process according to claim 1,2, 3, 4, 5, 6, 7, or 8, characterized in that the analysis detects stresses and that the phoneme string is extended in said places with stress symbols for different stress values to form a symbol string, the series of statistical values signal being reproduced unshortened or shortened according to the stress symbol when the microsegments are concatenated.
10. The speech-synthesis process according to claim 8 or 9, characterized in that provision is made for 5 levels of shortening by markings on the series of statistical values signal of the microsegments.
11. The speech-synthesis process according to claims 8 and 10, characterized in that the lengthening of the playback duration for phrase-final syllables takes place with closed syllables starting with the second microsegment of the vowel by increasing the shortening level for a longer playback duration in each case by one step, and with open syllables for the second microsegment of the vowel by increasing the shortening level for a longer playback duration by two steps.
12. The speech-synthesis process according to any one of the preceding claims, characterized in that the analysis allocates intonations and that the phoneme string is extended in said places with intonation symbols to form a symbol string, a fundamental frequency change of certain components of the periods being carried out on the intonation symbols in the time domain when the microsegments are concatenated.
13. The speech-synthesis process according to claim 12, characterized in that for reducing the fundamental frequency, defined sample values are added, or for increasing the fundamental frequency sample values are skipped in the open phase of the oscillation period of the vocal cords.
14. The speech-synthesis process according to claim 8,9, 10 11, 12, or 13, characterized in that the symbol string is converted into a microsegment string representing the sequence of microsegments and their modifications, taking into account the sequence phonemes and symbols.
15. The speech-synthesis process according to any one of the preceding claims, characterized in that the microsegments start with the first sample value after the first positive zero crossing and end with the last sample value before the last positive zero crossing.

Images