Method for modifying speech speed
3 claims: 3 independent, 0 dependent
- 1Method of modifying the speed of voice signals, in particular digitized voice signals, in which - an analog voice signal is digitized, thereby producing a digitized voice signal that is stored in a memory,- a factor α is defined by which the voice signal is lengthened or shortened,- a window function having a first, rising section of length N, a second, constant section of length L+N directly adjoining the first section and a third, falling section directly adjoining the second section is defined, wherein, if the first, rising section of a window overlaps the third, falling section of another window and the two sections are added in the overlap region, the result amounts to one, which corresponds to the value of the second section of the window function,- segments of a defined length L+N are extracted from the digitized, stored voice signal at irregular intervals of mean length αL,- said segments extracted from the digitized, stored voice signal are weighted with the window function in the time domain,- the weighted segments, each offset by a defined number of sample values L, are added, which shortens the voice signal thus produced for α > 1 and lengthens it for α < 1,- the segment extracted successively at the points of extraction of the segments from the digitized voice signal is compared there with the subsequently extracted segment, likewise weighted with the window function, for similarity,characterized- in that, for the purpose of rapidly comparing the similarity of the segments, only the N-value-long, third section, weighted with the falling window section, of the segment is compared with each first section, weighted with the rising N-value-long window section, of the subsequently extracted segment,- in that said segments, are added in an offset manner with respect to one another in such a way that the similarity of the two segment sections becomes maximal,- in that, to calculate the similarity, a correlation is used as a measure thereof. Method of modifying the speed of voice signals, in particular digitized voice signals, in which - an analog voice signal is digitized, thereby producing a digitized voice signal that is stored in a memory,- a factor α is defined by which the voice signal is lengthened or shortened,- a window function having a first, rising section of length N, a second, constant section of length L+N directly adjoining the first section and a third, falling section directly adjoining the second section is defined, wherein, if the first, rising section of a window overlaps the third, falling section of another window and the two sections are added in the overlap region, the result amounts to one, which corresponds to the value of the second section of the window function,- segments of a defined length L+N are extracted from the digitized, stored voice signal at irregular intervals of mean length αL,- said segments extracted from the digitized, stored voice signal are weighted with the window function in the time domain,- the weighted segments, each offset by a defined number of sample values L, are added, which shortens the voice signal thus produced for α > 1 and lengthens it for α < 1,- the segment extracted successively at the points of extraction of the segments from the digitized voice signal is compared there with the subsequently extracted segment, likewise weighted with the window function, for similarity,characterized- in that, for the purpose of rapidly comparing the similarity of the segments, only the N-value-long, third section, weighted with the falling window section, of the segment is compared with each first section, weighted with the rising N-value-long window section, of the subsequently extracted segment,- in that said segments, are added in an offset manner with respect to one another in such a way that the similarity of the two segment sections becomes maximal,- in that, to calculate the similarity, a correlation is used as a measure thereof. Procédé pour modifier la vitesse de signaux vocaux, notamment de signaux vocaux numérisés, selon lequel - un signal vocal analogique est numérisé, ce qui fait apparaître un signal vocal numérisé qui est mémorisé dans une mémoire,- un facteur α est défini, facteur avec lequel le signal vocal est allongé ou raccourci,- une fonction fenêtre comportant une première section montante de longueur N, une seconde section constante de longueur L-N, qui se raccorde directement à la première section, une troisième section' retombante, qui se raccorde à la seconde section, est définie, auquel cas lors d'une superposition de la première section montante d'une fenêtre avec la troisième section retombante d'une autre fenêtre et lors d'une addition des deux sections dans la zone de chevauchement, on obtient le résultat un, qui correspond à la valeur de la seconde section de la fonction fenêtre,- αL segments ayant une longueur définie L+N sont prélevés du signal vocal numérisé et mémorisé, à des intervalles irréguliers ayant une longueur moyenne,- ces segments prélevés du signal numérisé et mémorisé sont pondérés avec la fonction fenêtre dans le domaine temporel,- les segments pondérés sont additionnés en étant décalés respectivement d'un nombre défini de valeurs d'échantillonnage L, ce qui a pour effet que le signal vocal ainsi obtenu est raccourci pour α > 1 et est allongé pour α < 1,- le segment prélevé dans le signal vocal numérisé, et pondéré avec la fonction fenêtre est comparé, et ce successivement aux emplacements du prélèvement des segments à partir du signal vocal numérisé, au segment prélevé ensuite, également pondéré avec la fonction fenêtre, selon des aspects de similitude,caractérisé en ce- que pour la comparaison rapide de la similitude des segments, seule la troisième section du segment, qui possède une longueur de N valeurs et est pondérée par la section fenêtre retombante du segment est comparée à la section montante d'une longueur de N valeurs, qui est pondérée par la section fenêtre, du segment prélevé ensuite,- que ces segments sont additionnés en étant décalés les uns par rapport aux autres de telle sorte que la similitude des sections de segments devienne maximale, et- que pour le calcul de la similitude, on utilise une corrélation en tant que mesure de cette similitude. Verfahren zur Geschwindigkeitsmodifikation von Sprachsignalen, insbesondere digitalisierten Sprachsignalen, bei dem - ein analoges Sprachsignal digitalisiert wird, wodurch ein digitalisiertes Sprachsignal entsteht, welches in einem Speicher gespeichert wird,- ein Faktor α definiert wird, um welchen das Sprachsignal verlängert oder verkürzt wird,- eine Fensterfunktion mit einem ersten steigenden Abschnitt der Länge N, einem zweiten, sich direkt ah den ersten Abschnitt anschließenden, konstanten Abschnitt der Länge L-N und einem dritten, sich direkt an den zweiten Abschnitt anschließenden, fallenden Abschnitt definiert wird, wobei bei einer Überlagerung des ersten steigenden Abschnittes eines Fensters mit dem dritten fallenden Abschnitt eines anderen Fensters und einer Addition beider Abschnitte im Überlappungsbereich sich das Ergebnis eins ergibt, was dem Wert des zweiten Abschnittes der Fensterfunktion entspricht,- aus dem digitalisierten, gespeicherten Sprachsignal in unregelmäßigen Abständen einer mittleren Länge αL Segmente einer definierten Länge L+N entnommen werden,- diese, aus dem digitalisierten, gespeicherten Sprachsignal entnommenen Segmente mit der Fensterfunktion im Zeitbereich gewichtet werden,- die gewichteten Segmente jeweils um eine definierte Anzahl von Abtastwerten L versetzt aufaddiert werden, wodurch das so entstehende Sprachsignal für α > 1 verkürzt und für α < 1 verlängert wird,- nacheinander an den Stetten der Entnahme der Segmente aus dem digitalisierten Sprachsignal das dort entnommene, mit der Fensterfunktion gewichtete Segment mit dem nachfolgend entnommenen, ebenfalls mit der Fensterfunktion gewichteten Segment unter Ähnlichkeitsaspekten verglichen wird, dadurch gekennzeichnet,- dass zum schnellen Vergleich der Ähnlichkeit der Segmente lediglich der N Werte lange dritte, mit dem fallenden Fensterabschnitt gewichtete Abschnitt des Segmentes mit dem jeweils ersten, mit dem steigenden N Werte langen Fensterabschnitt gewichteten Abschnitt des nachfolgend entnommenen Segmentes verglichen wird,- dass diese Segmente so zueinander versetzt aufaddiert werden daß die Ähnlichkeit der beiden Segmentalschnitte maximal wird, und- dass zur Berechnung der Ähnlichkeit, als deren Maß, eine Korrelation verwendet wird.
- 2Method according to Claim 1, characterized in that the similarity of the two compared segment sections is a maximum if a maximization of the similarity measure is performed in relation to the displacement with respect to one another. Method according to Claim 1, characterized in that the similarity of the two compared segment sections is a maximum if a maximization of the similarity measure is performed in relation to the displacement with respect to one another. Procédé selon la revendication 1, caractérisé en ce que- la similitude des deux sections comparées de segment devient maximale lorsqu'on rend maximum le degré de similitude par rapport au décalage réciproque des segments. Verfahren nach Anspruch 1, dadurch gekennzeichnet, dass die Ähnlichkeit beider verglichener Segmentabschnitte maximal wird, wenn eine Maximierung des Ähnlichkeitsmaßes in Bezug zur Verschiebung zueinander durchgeführt wird.
- 3Method of modifying the speed of voice signals, in particular digitized voice signals in which - an analog voice signal is digitized, thereby producing a digitized voice signal that is stored in a memory,- a factor α is defined by which the voice signal is lengthened or shortened,- a window function having a first, rising section of length N, a second, constant section of length L+N directly adjoining the first section and a third, falling section directly adjoining the second section is defined, wherein, if the first, rising section of a window overlaps the third, falling section of another window and the two sections are added in the overlap region, the result amounts to one, which corresponds to the value of the second section of the window function,- segments of a length L+N are extracted from the digitized, stored voice signal at irregular intervals of mean length αL,- said segments extracted from the digitized, stored voice signal are weighted with the window function in the time domain,- the weighted segments, each offset by a defined number of sample values L, are added, which shortens the voice signal thus produced for α > 1 and lengthens it for α < 1,- the segment extracted successively at the points of extraction of the segments from the digitized voice signal is compared there in each case with the segment of the lengthened or shortened voice signal that represents said extracted segment,characterized- in that, for the purpose of rapidly comparing the deviation of the lengthened or shortened voice signal from the digitized voice signal, only the N-value-long, third section of the segment extracted last is used as reference,- in that the segments extracted are added in an offset manner to one another in such a way that the deviation determined is a minimum, and- the relative error or the absolute square error is used as a measure of the deviation. Method of modifying the speed of voice signals, in particular digitized voice signals in which - an analog voice signal is digitized, thereby producing a digitized voice signal that is stored in a memory,- a factor α is defined by which the voice signal is lengthened or shortened,- a window function having a first, rising section of length N, a second, constant section of length L+N directly adjoining the first section and a third, falling section directly adjoining the second section is defined, wherein, if the first, rising section of a window overlaps the third, falling section of another window and the two sections are added in the overlap region, the result amounts to one, which corresponds to the value of the second section of the window function,- segments of a length L+N are extracted from the digitized, stored voice signal at irregular intervals of mean length αL,- said segments extracted from the digitized, stored voice signal are weighted with the window function in the time domain,- the weighted segments, each offset by a defined number of sample values L, are added, which shortens the voice signal thus produced for α > 1 and lengthens it for α < 1,- the segment extracted successively at the points of extraction of the segments from the digitized voice signal is compared there in each case with the segment of the lengthened or shortened voice signal that represents said extracted segment,characterized- in that, for the purpose of rapidly comparing the deviation of the lengthened or shortened voice signal from the digitized voice signal, only the N-value-long, third section of the segment extracted last is used as reference,- in that the segments extracted are added in an offset manner to one another in such a way that the deviation determined is a minimum, and- the relative error or the absolute square error is used as a measure of the deviation. Procédé pour modifier la vitesse de signaux vocaux, notamment de signaux vocaux numérisés, selon lequel un signal vocal analogique est numérisé, ce qui fait apparaître un signal vocal numérisé qui est mémorisé dans une mémoire, - un facteur α est défini, facteur avec lequel le signal vocal est allongé ou raccourci,- une fonction fenêtre comportant une première section montante de longueur N, une seconde section constante de longueur L-N, qui se raccorde directement à la première section, et une troisième section retombante, qui se raccorde à la seconde section, est définie, auquel cas lors d'une superposition de la première section montante d'une fenêtre avec la troisième section retombante d'une autre fenêtre et lors d'une addition des deux sections dans la zone de chevauchement, on obtient le résultat un, qui correspond à la valeur de la seconde section de la fonction fenêtre,- αL segments ayant une longueur définie L+N sont prélevés du signal vocal numérisé et mémorisé, à des intervalles irréguliers ayant une longueur moyenne,- ces segments prélevés du signal numérisé et mémorisé sont pondérés avec la fonction fenêtre dans le domaine temporel,- les segments pondérés sont additionnés en étant décalés respectivement d'un nombre défini de valeurs d'échantillonnage L, ce qui a pour effet que le signal vocal ainsi obtenu est raccourci pour α > 1 et est allongé pour α < 1,- le segment prélevé dans le signal vocal numérisé, et pondéré avec la fonction fenêtre est comparé, et ce successivement aux emplacements de prélèvement des segments à partir du signal vocal numérisé, au segment du signal vocal allongé ou raccourci, qui représente ce segment prélevé,caractérisé en ce- que pour la comparaison rapide de l'écart entre le signal vocal allongé ou le signal vocal raccourci par rapport au signal vocal numérisé, on utilise comme référence uniquement la troisième section, d'une grandeur de N valeurs, du signal prélevé en dernier,- qu'on additionne les segments prélevés d'une manière décalée entre eux de telle sorte que l'écart déterminé est minimum, et- qu'on utilise comme mesure de l'écart l'erreur relative ou l'erreur quadratique absolue. Verfahren zur Geschwindigkeitsmodifikation von Sprachsignalen, insbesondere digitalisierten Sprachsignalen, bei dem - ein analoges Sprachsignal digitalisiert wird, wodurch ein digitalisiertes Sprachsignal entsteht, welches in einem Speicher gespeichert wird,- ein Faktor α definiert wird, um welchen das Sprachsignal verlängert oder verkürzt wird,- eine Fensterfunktion mit einem ersten steigenden Abschnitt der Länge N, einem zweiten, sich direkt an den ersten Abschnitt anschließenden, konstanten Abschnitt der Länge L-N und einem dritten, sich direkt an den zweiten Abschnitt anschließenden, fallenden Abschnitt definiert wird, wobei bei einer Überlagerung des ersten steigenden Abschnittes eines Fensters mit dem dritten fallenden Abschnitt eines anderen Fensters und einer Addition beider Abschnitte im Überlappungsbereich sich das Ergebnis eins ergibt, was dem Wert des zweiten Abschnittes der Fensterfunktion entspricht,- aus dem digitalisierten, gespeicherten Sprachsignal in unregelmäßigen Abständen einer mittleren Länge αL Segmente einer Länge L+N entnommen werden,- diese, aus dem digitalisierten, gespeicherten Sprachsignal entnommenen Segmente mit der Fensterfunktion im Zeitbereich gewichtet werden,- die gewichteten Segmente jeweils um eine definierte Anzahl von Abtastwerten L versetzt aufaddiert werden, wodurch das so entstehende Sprachsignal für α > 1 verkürzt und für α < 1 verlängert wird,- nacheinander an den Stellen der Entnahme der Segmente aus dem digitalisierten Sprachsignal, jeweils das dort entnommene Segment mit dem Segment des verlängerten oder verkürzten Sprachsignals, welches dieses entnommene Segment repräsentiert, verglichen wird, dadurch gekennzeichnet,- dass zum schnellen Vergleich der Abweichung des verlängerten oder verkürzten Sprachsignals vom digitalisierten Sprachsignal lediglich der N Werte lange dritte Abschnitt des zuletzt entnommenen Segmentes als Referenz herangezogen wird,- dass die entnommenen Segmente so zueinander versetzt aufaddiert werden, daß die ermittelte Abweichung minimal ist und- dass als Maß für die Abweichung der relative Fehler oder der absolute quadratische Fehler herangezogen wird.
Independent claims3
22 paragraphs, as filed
The invention relates to a method for speed modification of speech signals in the time domain, in particular an efficient overlap-add method.
In various areas of the processing of speech and audio signals, a change in the playback speed of these signals is desired, if possible without impairing their naturalness and, in the case of speech, their intelligibility. From a technical point of view, this goal of preserving the sound character can be formulated as follows: despite a modification of the time scale of these signals, their short-term spectral properties should remain unchanged. In particular, this means for speech signals that fundamental frequency and formants must be preserved in the speed modification.
The time compression or time stretching of audio signals is used in studios, for example, with the aim of trimming commercials to the intended length. Also in dictation is the adaptation of the playback speed to the needs or Skills of the typist of importance. Another application is the real time transmission of voice signals, where variable delay data packets arrive at the receiver. By applying the speed modification one can keep the over-all delay on average lower than the worst-case delay of the transmission path, without a data packet arriving too late would lead to dropouts or other similarly disturbing effects. In addition to the desire for the highest possible sound quality, the following additional requirements for the method arise for many applications:
A cost real-time realization must be achievable, and it must be possible at runtime if possible stepless change of the speed modification factor. Without a doubt, it is also advantageous if the algorithm manages without an always erroneous pitch estimate.
From Method for Time or Frequency Compression Expansion of Speed, by G. Fairbaks and R. P. Jaeger, Inst. of Radio Engineers Trans. on Audio, Vol. AU-2, No. 1, pp. 7-12, Jan. 1954, are first investigations for speech signal compression or Speech signal stretching known. Frequency domain methods have frequently been used since then - obvious, since, as mentioned above, the short-term spectral characteristics of the speech signal should be preserved. Since the mid-eighties, comparatively simple time-lapse overlap-add methods have been known with which very well-timed voice signals can be generated.
In "Signal Estimation from Modified Short-Time Fourier Transform", by D. W. Griffin, in IEEE Trans. Acoust., Speech, Signal Processing, Vol. ASSP-32, no. 2, pp. 236-242, Apr. In 1984, Griffin and Lim report experiments with a very elaborate iterative phase determination. In turn, the publication of S. takes on this approach. Roucos and A. M. Wilgu's High Quality Time-Scale Modification for Speech, IEEE Proc. Int. Conf. Acoust., Speech, Signal Processing, pp. 493-496, 1985, which propose a time domain method which generates time scaled speech signals by means of an overlap-add approach. In this so-called SOLA method (SOLA = Synchronized OverLap Add), the sections taken at regular intervals from the original signal are synchronized by shifting in front of the respectively corresponding windowing and addition in the target signal. This corresponds in a broader sense to the phase optimization, as it is carried out in the frequency domain method. Closely related to the SOLA algorithm is the so-called WSOLA method (WSOLA = Waveform Similarity OverLap - Add), which is W. Verhelst and M. Roelands in "An Overlap-Add Technique Based on Waveform Similarity (WSOLA) for High Quality Time-Scale Modification of Speed," IEEE Proc. Int. Conf. Acoust., Speech, Signal Processing, pp. 554-567, 1993, and "Waveform Simulatority-Based Overlap-Add (WSOLA) for Time-Scale Modification of Speech: Structures and Evaluation", Int. Conf. on Speech Communication and Technology, pp. 337-340, 1993. The main difference between these two approaches is the synchronization, which is carried out in the WSOLA method by offset removal of segments from the original signal, which in relation to the SOLA principle, especially reduces the overhead.
The object of the invention is to provide a method for speed modification of speech signals in the time domain, which works very efficiently.
This object is solved by the features of claims 1 and 3. Advantageous embodiments of the invention are given in the following description.
The generation of the time-scaled with the factor α version y (k) of a speech signal x (k) is carried out according to the synthesis<maths id="math0001" num=""><img file="EP0865026B1_D0001.tif" /></maths> with a window function<maths id="math0002" num=""><img file="EP0865026B1_D0002.tif" /></maths>
The function v (k) defined here for k = 0,..., N-1 is expediently between its extremes v (0) = ε<sub>0</sub> with 0 <ε<sub>0</sub><< 1 and v (N-1) = 1-ε<sub>1</sub> with 0 <ε<sub>1</sub><< 1 monotonously growing.
The specified w (k) definition ensures that the necessary condition for meaningful overlap-add<maths id="math0003" num=""><img file="EP0865026B1_D0003.tif" /></maths> is satisfied.
The displacement variable Δ contained in the above equation of synthesis<sub>λ</sub> is for the purpose of the mentioned synchronization of a "tolerance range" -Δ<sub>Max</sub>, ..., Δ<sub>Max</sub> to determine.
The basic procedure is as follows: Apart from a synchronization-dependent "jitter", segments of length L + N are taken from the original signal x (k) and, after weighting with w (k), added by L samples. The thus-obtained signal y (k) is accelerated by a factor α with respect to x (k), that is, an utterance of K sample length contained in the original signal x (k) by this procedure on a y (k) portion the length K / α imaged, so shortened and thus accelerated in the playback for α> 1, or is extended, that is slowed down, when α <1.
The synchronization of the overlapping sections is of great importance for the resulting sound quality. For this purpose, the following approach is used: During the execution of the method, for each segment taken from the signal x (k), the section of x (k) offset by L samples may be regarded as the "ideal segment" for the next step Overlap add operation again the original signal x (k) would reproduce. However, the desired time scaling now requires that for the overlap-add synthesis i. a. another section of x (k) offset from the "ideal segment" is selected. The best possible synchronization is given if the section used for the overlap add operation has the greatest possible similarity ("Waveform Similarity") to the "ideal segment".
As a criterion for the similarity of said segments offer different dimensions. For example, the use of the correlation coefficient is obvious. While W. Verhelst and M. Roelands in "An Overlap-Add Technique Based on Waveform Similarity (WSOLA) for High Quality Time-Scale Modification of Speed," in IEEE Proc. Int. Conf. Acoust., Speech, Signal Processing, pp. 554-557, 1993, and "Waveform Similarity Based Overlap-Add (WSOLA) for Time-Scale Modification of Speech: Structures and Evaluation" in Int. Conf. on Speech Communication and Technology, pp. 337-340, 1993, have used the complete segment of length L + N for the evaluation of the similarity measure, it seems perfectly sufficient to restrict the calculation to the range of N samples in which the segments actually overlap.
For further illustrations, it is helpful to introduce the following vector infotation: The x values long section of the "ideal segment" in which the overlap with the new segment to be determined will be denoted by x, the first N values of the shifted segment by x<sub>q</sub>, The weighting of this section with the rising edge of the window is represented by multiplying this vector by a diagonal matrix V that matches the values v (0), ... v (N-1) is occupied. Accordingly, the weighting of the ideal segment section x with the falling edge of the window is multiplied by<b>1</b> - V shown, where <b>1</b> denotes the N × N unit matrix. The resulting in the critical overlap area from the overlap-add synthesis y (k) section is so<maths id="math0004" num=""><math display="block"><mrow><msub><mrow><mtext>y = (1-V) x + Vx</mtext></mrow><mrow><mtext>q</mtext></mrow></msub></mrow></math><img file="EP0865026B1_D0004.tif" /></maths>
For example, a cross-correlated calculation can now be used as a measure of the similarity of the components involved<maths id="math0005" num=""><math display="block"><mrow><msub><mrow><mtext>C</mtext></mrow><mrow><mtext>δ</mtext></mrow></msub><msup><mrow><mtext> = x</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><msup><mrow><mtext>(1-V)</mtext></mrow><mrow><mtext>T</mtext></mrow></msup><msub><mrow><mtext> V x</mtext></mrow><mrow><mtext>q</mtext></mrow></msub></mrow></math><img file="EP0865026B1_D0005.tif" /></maths> specify. The maximization of this expression with respect to x<sub>q</sub> finding displacement δ ∈ {-Δ<sub>Max</sub>, ..., Δ<sub>Max</sub>} returns the optimal shift Δ for the considered segment in terms of the assumed similarity measure<sub>λ</sub>,
The calculation of C<sub>δ</sub> requires all L samples 2N multiplies for the precomputation of the expression x<sup>T</sup>(<b>1</b>-V)<sup>T</sup>V and then (2Δ<sub>Max</sub>+1) N multiplications and additions.
This is opposite W. Verhelst and M. Roelands in "An Overlap-Add Technique Based on Waveform Similarity (WSOLA) for High Quality Time-Scale Modification of Speed," in IEEE Proc. Int. Conf. Acoust., Speech, Signal Processing, pp. 554-557, 1993, and "Waveform Similarity Based Overlap-Add (WSOLA) for Time-Scale Modification of Speech: Structures and Evaluation" in Int. Conf. on Speech Communication and Technology, pp. 337-340, 1993, a cost reduction by a factor of two, which even increases for L> N. Limiting the similarity calculation to the area of overlap has no negative impact on the quality of time-scaled speech samples.
Another approach to synchronization is to minimize the error between the synthesized signal y and the original signal x instead of maximizing the waveform similarity. A simple arbitrary choice is the square expression for this error<maths id="math0006" num=""><math display="block"><mrow><msub><mrow><mtext>e</mtext></mrow><mrow><mtext>δ</mtext></mrow></msub><msup><mrow><mtext> = ∥x - y∥</mtext></mrow><mrow><mtext>2</mtext></mrow></msup></mrow></math><img file="EP0865026B1_D0006.tif" /></maths> to be set.
Neglecting the preliminary calculations, the value for the evaluation of E<sub>δ</sub> incurred effort on (2Δ<sub>Max</sub>+1) 4N DSP operations every L samples. This is understood to mean those operations that a signal processor with common architecture can process in one step.
Another approach is to use the relative error instead of the absolute error<maths id="math0007" num=""><math display="block"><mrow><mtext mathvariant="italic">R</mtext><mtext>∂ = </mtext><mfrac><mrow><mtext>∥</mtext><mtext mathvariant="italic">x</mtext><mtext> - </mtext><mtext mathvariant="italic">y</mtext><msup><mrow><mtext>∥</mtext></mrow><mrow><mtext>2</mtext></mrow></msup></mrow><mrow><mtext>∥</mtext><mtext mathvariant="italic">y</mtext><msup><mrow><mtext>∥</mtext></mrow><mrow><mtext>2</mtext></mrow></msup></mrow></mfrac></mrow></math><img file="EP0865026B1_D0007.tif" /></maths> to minimize what can be interpreted as SNR maximization. (2Δ<sub>Max</sub>+1) 5N operations are required here before any overlap add operation.
7 sheets
Sheet 1 Sheet 2 Sheet 3 Sheet 4 Sheet 5 Sheet 6 Sheet 7
Every citation, both ways
| Document | Relation | Office |
|---|---|---|
| EP0427953A | Cites | European Patent Office (EPO) |
| EP0608833A | Cites | European Patent Office (EPO) |
| EP0726560A | Cites | European Patent Office (EPO) |
7 members in 3 offices
Priority claims5
| Document | Office | Kind | Date |
|---|---|---|---|
| 19710545 | Germany | A | |
| 19710545 | Germany | A | |
| 19710545 | Germany | – | |
| 19710545 | – | – | – |
| DE1997110545 | – | – | – |
Members7
| Document | Office | Kind | |
|---|---|---|---|
| DE19710545C1 | Germany | C1 | |
| EP0865026A2 | European Patent Office (EPO) | A2 | |
| EP0865026A3 | European Patent Office (EPO) | A3 | |
| EP0865026B1This record | European Patent Office (EPO) | B1 | |
| AT255763T | Austria | T | |
| ATE255763T1 | Austria | T1 | |
| DE59810302D1 | Germany | D1 |
44 legal events, as 5 offices reported them to INPADOC
Over the term
Point at a mark for the eventEvents
| Event | Code | Office | |
|---|---|---|---|
| Lapsed in a contracting state [announced via postgrant information from national office to epo]LapsedPG25 | PG25 | EP | |
| Lapsed in a contracting state [announced via postgrant information from national office to epo]LapsedPG25 | PG25 | EP | |
| Lapsed in a contracting state [announced via postgrant information from national office to epo]LapsedPG25 | PG25 | EP | |
| Notification of lapseLapsedST | ST | FR | |
| Gb: european patent ceased through non-payment of renewal feeCeasedGBPC | GBPC | EP | |
| Application deemed withdrawn, or ip right lapsed, due to non-payment of renewal feeWithdrawnR119 | R119 | DE | |
| Annual fee paid to national office [announced via postgrant information from national office to epo]GrantedPGFP | PGFP | EP | |
| Annual fee paid to national office [announced via postgrant information from national office to epo]GrantedPGFP | PGFP | EP | |
| Annual fee paid to national office [announced via postgrant information from national office to epo]GrantedPGFP | PGFP | EP | |
| Change of representativeR082 | R082 | DE | |
| Fee paymentPLFP | PLFP | FR | |
| Lapsed in a contracting state [announced via postgrant information from national office to epo]LapsedPG25 | PG25 | EP | |
| Lapsed in a contracting state [announced via postgrant information from national office to epo]LapsedPG25 | PG25 | EP | |
| Lapsed because of non-payment of the annual feeLapsedV1 | V1 | NL | |
| Annual fee paid to national office [announced via postgrant information from national office to epo]GrantedPGFP | PGFP | EP | |
| Annual fee paid to national office [announced via postgrant information from national office to epo]GrantedPGFP | PGFP | EP | |
| Nl: assignments of ep-patentsNLS | NLS | EP | |
| Transmission of propertyTP | TP | FR | |
| Amendments to the register in respect of changes of name or changes affecting rights (sect. 32/1977)732E | 732E | GB | |
| No opposition filedOpposition26N | 26N | EP | |
| No opposition filed within time limitOppositionORIGINAL CODE: 0009261PLBE | PLBE | EP | |
| Information on the status of an ep patent application or granted ep patentGrantedSTATUS: NO OPPOSITION FILED WITHIN TIME LIMITSTAA | STAA | EP | |
| Fr: translation filedET | ET | EP | |
| Corresponds to:REF | REF | EP | |
| Gb: translation of ep patent filed (gb section 77(6)(a)/1977)GBT | GBT | EP | |
| Designated contracting statesAK | AK | EP | |
| European patent grantedGrantedNOT ENGLISHFG4D | FG4D | GB | |
| (expected) grantORIGINAL CODE: 0009210GRAA | GRAA | EP | |
| Grant fee paidORIGINAL CODE: EPIDOSNIGR3GRAS | GRAS | EP | |
| Information provided on ipc code assigned before grantRIC1 | RIC1 | EP | |
| Information provided on ipc code assigned before grantRIC1 | RIC1 | EP | |
| Title (correction)METHOD FOR MODIFYING SPEECH SPEEDRTI1 | RTI1 | EP | |
| Despatch of communication of intention to grant a patentORIGINAL CODE: EPIDOS IGRAGRAH | GRAH | EP | |
| First examination report despatched17Q | 17Q | EP | |
| Information provided on ipc code assigned before grant7G 10L 21/04 ARIC1 | RIC1 | EP | |
| Party data changed (applicant data changed or rights of an application transferred)RAP1 | RAP1 | EP | |
| Designation fees paidAT DE FR GB NLAKX | AKX | EP | |
| Request for examination filed17P | 17P | EP | |
| Designated contracting statesAK | AK | EP | |
| Request for extension of the european patentAL;LT;LV;MK;RO;SIAX | AX | EP | |
| Search report despatchedORIGINAL CODE: 0009013PUAL | PUAL | EP | |
| Designated contracting statesAK | AK | EP | |
| Request for extension of the european patentAL;LT;LV;MK;RO;SIAX | AX | EP | |
| Public reference made under article 153(3) epc to a published international application that has entered the european phaseORIGINAL CODE: 0009012PUAI | PUAI | EP |
Numbers
- Publication
- 0865026
- Publication, DOCDB
- 0865026
- Publication, EPODOC
- EP0865026
- Application
- 98104455
- Application, DOCDB
- 98104455
- Application, EPODOC
- EP19980104455
Titles3
- German
- Effizientes Verfahren zur Geschwindigkeitsmodifikation von Sprachsignalen
- English
- Method for modifying speech speed
- French
- Méthode pour la modification du débit de parole
Classification
- CPC, 1
- G10L21/04
- IPC, 1
- G10L21 04
Designated states5
- Contracting states, 5
- Austria
- Germany
- France
- United Kingdom
- Netherlands (Kingdom of the)
