EP1048025B1 - Method for objective voice quality evaluation - Google Patents

Method for objective voice quality evaluation Download PDF

Info

Publication number
EP1048025B1
EP1048025B1 EP99942871A EP99942871A EP1048025B1 EP 1048025 B1 EP1048025 B1 EP 1048025B1 EP 99942871 A EP99942871 A EP 99942871A EP 99942871 A EP99942871 A EP 99942871A EP 1048025 B1 EP1048025 B1 EP 1048025B1
Authority
EP
European Patent Office
Prior art keywords
spectral
computed
speech signal
evaluated
signal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Lifetime
Application number
EP99942871A
Other languages
German (de)
French (fr)
Other versions
EP1048025A1 (en
Inventor
Jens Berger
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Deutsche Telekom AG
Original Assignee
Deutsche Telekom AG
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Deutsche Telekom AG filed Critical Deutsche Telekom AG
Publication of EP1048025A1 publication Critical patent/EP1048025A1/en
Application granted granted Critical
Publication of EP1048025B1 publication Critical patent/EP1048025B1/en
Anticipated expiration legal-status Critical
Expired - Lifetime legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
    • G10L25/69—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for evaluating synthetic or decoded voice signals
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use

Definitions

  • the invention relates to a method for instrumental ("objective") Speech quality determination, by comparing the properties of one evaluating speech signal with properties of a reference speech signal (undisturbed Signal) characteristic values for determining the speech quality (speech quality) can be derived.
  • Speech quality determinations of speech signals are generally made using auditory ("subjective") investigations with test persons.
  • the goal of instrumental ("objective") procedures for determining speech quality is it, from properties of the speech signal to be evaluated by means of suitable computing methods To determine characteristic values that determine the speech quality of the speech signal to be assessed describe without having to resort to judgments from test subjects.
  • the calculated characteristic values and the underlying method for instrumental Speech quality assessment is considered recognized if there is a high correlation Results of auditory comparative examinations is achieved.
  • the means of auditory Examinations of speech quality values thus represent the target values that to be achieved through instrumental procedures.
  • Known methods for instrumental speech quality determination are based on a comparison of a reference speech signal with the speech signal to be evaluated.
  • the reference speech signal and the speech signal to be evaluated are segmented into short time segments.
  • the spectral properties of the two signals are compared in these segments.
  • Various approaches and models are used to calculate the spectral short-term properties.
  • the signal intensity is calculated in frequency bands whose width increases with increasing center frequency. Examples of such frequency bands are the known third octave bands or frequency groups according to Zwicker (published in Zwicker, E .: “Psychoakustik", Berlin: Springer-Verlag, 1982).
  • the spectral intensity map calculated in this way for each period of time under consideration can be understood as a series of numerical values in which the number of individual values of the The number of frequency bands used corresponds to the numerical values even the calculated ones Represent intensity values and a continuous index of the frequency bands the order which describes numerical values.
  • the calculated intensities are to be evaluated Voice signal and reference voice signal compared in each band.
  • the difference both values, or the similarity of the two resulting spectral values Intensity maps, is the basis for the calculation of a quality value (Fig. 1).
  • a disadvantage of the methods known today in such cases is that a comparison differences between the speech signal to be evaluated with a reference speech signal between the two signal sections in the selected display level in the calculative quality parameter that does not or hardly at all - also in audible test perceptible - lead to qualitative impairment.
  • Frequency band limits and spectral deformations of the speech signal to be evaluated contributes only to a limited extent to a perceived qualitative impairment.
  • the invention has as its object the influence of spectral limits and Deformations of the speech signal to be assessed and of spectral shifts Short-term maxima before comparing the spectral properties of a test item Signal with a reference speech signal and the calculation of a quality value in to reduce instrumental procedures.
  • one is used in the invention described here generated spectral weighting function based on medium spectral envelopes, e.g. the average spectral power density, of the speech signal to be evaluated and Reference speech signal based. This also enables the use of the method non-linear and time-variant transmission.
  • the evaluation function a (f) can weight the weighting function W T (f) differently over the effective range, in the simplest case it is constant 1.
  • the spectral weighting function W T (f) calculated in this way approximates the mean spectral envelopes of the speech signal and the reference speech signal to be evaluated, so that differences between the two spectral envelopes only have a reduced effect on the calculated quality value.
  • the spectral weighting function W T (f) can be applied to the reference speech signal.
  • the average spectral power density of the reference speech signal is approximated to the signal to be evaluated (FIG. 2a).
  • the spectral weighting function can be inverted to the signal to be evaluated be applied. This is equalized and, with regard to its middle one spectral power density, approximated to the reference speech signal (Fig. 2b).
  • Another part of the invention relates to the correction of displacements short-term spectral maxima caused by the transmission systems.
  • the intensity is integrated in frequency bands for each time period.
  • the result is one Series of intensity values for each spectral representation of a signal section, where each individual value represents the intensity in a frequency band.
  • the shifts Short-term spectral maxima can lead to deviating calculated intensities in the frequency bands of the reference speech signal and the speech signal to be evaluated.
  • variable band limits for calculating the spectral intensity mapping is not only limited to the signal in which the described spectral weighting function W T (f) is also used, but can also be applied to the other signal and even to both signals. (see FIGS. 2a and 2b).
  • a special embodiment shows an implementation according to FIG. 3, which is called TOSQA (Telecommunication Objective Speech Quality Assessment). This is done advanced preprocessing of the reference speech signal.
  • TOSQA Telecommunication Objective Speech Quality Assessment
  • Speech pauses are recognized by a speech pause recognizer and do not go into that Quality measure.
  • the reference speech signal and is also filtered evaluating speech signal with a bandpass 300 ... 3400 Hz and a filtering on the Frequency response of a telephone handset.
  • the integration of the spectral power density takes place in frequency groups, which are the basis for the calculation of specific loudness represent.
  • the calculated loudness patterns are in addition to the general approach supplemented by an error evaluation function.
  • the calculated quality value is over a Average of the correlation coefficients of the specific loudness for each considered short time segment over the number of evaluated language segments formed.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)
  • Machine Translation (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

In a method for determining speech quality using an objective measure, in order to enhance prediction reliability of the evaluated quality parameters, distortions of the mean spectral envelope are extensively corrected with a weighting function W<SUB>T</SUB>(f) before comparing spectral properties. Additionally, the fixed band limits for integration of spectral power density are suppressed and other band limits are searched for instead in a predetermined optimization area in which the resulting spectral intensity representations of the voice signal to be evaluated and the reference voice signal have maximum similarity. The solutions described can supplement known methods and can be incorporated into their structures.

Description

VorbemerkungPreliminary note

Die Erfindung bezieht sich auf ein Verfahren zur instrumentellen ("objektiven") Sprachqualitätsbestimmung, bei dem durch Vergleich von Eigenschaften eines zu bewertenden Sprachsignals mit Eigenschaften eines Referenzsprachsignals (ungestörtes Signal) Kennwerte zur Bestimmung der Sprachqualität (Sprachgüte) abgeleitet werden.The invention relates to a method for instrumental ("objective") Speech quality determination, by comparing the properties of one evaluating speech signal with properties of a reference speech signal (undisturbed Signal) characteristic values for determining the speech quality (speech quality) can be derived.

Sprachqualitätsbestimmungen von Sprachsignalen werden in der Regel mittels auditiver ("subjektiver") Untersuchungen mit Versuchspersonen vorgenommen.Speech quality determinations of speech signals are generally made using auditory ("subjective") investigations with test persons.

Das Ziel von instrumentellen ("objektiven") Verfahren zur Sprachqualitätsbestimmung ist es, aus Eigenschaften des zu bewertenden Sprachsignals mittels geeigneter Rechenverfahren Kennwerte zu ermitteln, die die Sprachqualität des zu bewertenden Sprachsignals beschreiben, ohne auf Urteile von Versuchspersonen zurückgreifen zu müssen.The goal of instrumental ("objective") procedures for determining speech quality is it, from properties of the speech signal to be evaluated by means of suitable computing methods To determine characteristic values that determine the speech quality of the speech signal to be assessed describe without having to resort to judgments from test subjects.

Die berechneten Kennwerte und das zugrunde gelegte Verfahren zur instrumentellen Sprachqualitätsbestimmung gelten als anerkannt, wenn eine hohe Korrelation zu Ergebnissen auditiver Vergleichsuntersuchungen erreicht wird. Die mittels auditiver Untersuchungen gewonnenen Sprachqualitätswerte stellen somit die Zielwerte dar, die durch instrumentelle Verfahren erreicht werden sollen.The calculated characteristic values and the underlying method for instrumental Speech quality assessment is considered recognized if there is a high correlation Results of auditory comparative examinations is achieved. The means of auditory Examinations of speech quality values thus represent the target values that to be achieved through instrumental procedures.

Stand der TechnikState of the art

Bekannte Verfahren zur instrumentellen Sprachqualitätsbestimmung beruhen auf einem Vergleich eines Referenzsprachsignals mit dem zu bewertenden Sprachsignal. Dabei werden das Referenzsprachsignal und das zu bewertendes Sprachsignal in kurze Zeitabschnitte segmentiert. In diesen Segmenten werden die spektralen Eigenschaften der beiden Signale verglichen.
Für die Berechnung der spektralen Kurzzeiteigenschaften kommen verschiedene Ansätze und Modelle zur Anwendung. In der Regel erfolgt die Berechnung der Signalintensität in Frequenzbändern, deren Breite mit zunehmender Mittenfrequenz größer wird. Beispiele für solche Frequenzbänder sind die bekannten Terzbänder oder Frequenzgruppen nach Zwicker (veröffentlicht in Zwicker, E.: "Psychoakustik", Berlin: Springer-Verlag, 1982).Known methods for instrumental speech quality determination are based on a comparison of a reference speech signal with the speech signal to be evaluated. The reference speech signal and the speech signal to be evaluated are segmented into short time segments. The spectral properties of the two signals are compared in these segments.
Various approaches and models are used to calculate the spectral short-term properties. As a rule, the signal intensity is calculated in frequency bands whose width increases with increasing center frequency. Examples of such frequency bands are the known third octave bands or frequency groups according to Zwicker (published in Zwicker, E .: "Psychoakustik", Berlin: Springer-Verlag, 1982).

Die derart berechnete spektrale Intensitätsabbildung für jeden betrachteten Zeitabschnitt läßt sich als Reihe von Zahlenwerten auffassen, in der die Anzahl der Einzelwerte der Anzahl der verwendeten Frequenzbänder entspricht, die Zahlenwerte selbst die berechneten Intensitätswerte darstellen und ein fortlaufender Index der Frequenzbänder die Reihenfolge der Zahlenwerte beschreibt.The spectral intensity map calculated in this way for each period of time under consideration can be understood as a series of numerical values in which the number of individual values of the The number of frequency bands used corresponds to the numerical values even the calculated ones Represent intensity values and a continuous index of the frequency bands the order which describes numerical values.

Bei den derzeit bekannten Verfahren zur instrumentellen Sprachqualitätsbestimmung werden die Grenzen der benutzten Frequenzbänder auf der Frequenzachse konstant gehalten.In the currently known methods for instrumental language quality determination the limits of the frequency bands used on the frequency axis become constant held.

In jedem betrachteten Zeitsegment werden die berechneten Intensitäten von zu bewertenden Sprachsignal und Referenzsprachsignal in jedem Band miteinander verglichen. Die Differenz beider Werte, bzw. die Ähnlichkeit der beiden entstehenden spektralen Intensitätsabbildungen, stellt die Grundlage für die Berechnung eines Qualitätswertes dar (Fig. 1).In each time segment considered, the calculated intensities are to be evaluated Voice signal and reference voice signal compared in each band. The difference both values, or the similarity of the two resulting spectral values Intensity maps, is the basis for the calculation of a quality value (Fig. 1).

Solche Verfahren wurden insbesondere für die qualitative Bewertung der Sprache in der Telefonieanwendung entwickelt. Beispiele hierfür sind US-A-5,621,854 (HOLLIER MICHAEL P), die als nächstliegender Stand der Technik zitiert wird, und die Veröffentlichungen:

  • "A perceptual speech-quality measure based on a psychacoustic sound representation" (Beerends, J. G.; Stemerdink, J. A., J. Audio Eng. Soc. 42(1994)3, S. 115-123)
  • "Auditory distortion measure for speech coding" (Wang, S; Sekey, A.; Gersho, A.: IEEE Proc. Int. Conf. acoust., speech and signalprocessing (1991), S.493-496).
    • Such methods were developed especially for the qualitative assessment of language in the telephony application. Examples include US-A-5,621,854 (HOLLIER MICHAEL P), which is cited as the closest prior art, and the publications:
  • "A perceptual speech-quality measure based on a psychacoustic sound representation" (Beerends, JG; Stemerdink, JA, J. Audio Eng. Soc. 42 (1994) 3, pp. 115-123)
  • "Auditory distortion measure for speech coding" (Wang, S; Sekey, A .; Gersho, A .: IEEE Proc. Int. Conf. Acoust., Speech and signal processing (1991), S.493-496).

    • Der derzeit gültige ITU-T Standard P.861 beschreibt ebenfalls ein derartiges Verfahren: "Objective quality measurement of telephone-band speech codecs" (ITU-T Rec. P.861, Genf 1996). The currently valid ITU-T standard P.861 also describes such a procedure: "Objective quality measurement of telephone-band speech codecs" (ITU-T Rec. P.861, Geneva 1996).

    Nachteile bekannter instrumenteller SprachqualitätsmeßverfahrenDisadvantages of known instrumental speech quality measurement methods

    Der Einsatz von bekannten Verfahren zur instrumentellen Sprachqualitätsbestimmung scheitert an der Zuverlässigkeit der berechneten Qualitätswerte für bestimmte zu bewertende Signaleigenschaften. Insbesondere bei Beeinträchtigungen im zu bewertenden Sprachsignal, wie sie z.B. durch Sprachcodierverfahren mit niedrigen Bitraten oder Kombinationen von unterschiedlichen Störungen hervorgerufen werden, liefern derzeit bekannte Verfahren nur unsichere Qualitätswerte.The use of known methods for instrumental language quality determination fails due to the reliability of the calculated quality values for certain ones evaluating signal properties. Especially in the case of impairments in the Speech signal, e.g. through speech coding methods with low bit rates or Combinations of different disorders are currently being delivered known methods only unsafe quality values.

    Nachteilig bei den heute bekannten Verfahren ist in solchen Fällen, daß bei einem Vergleich zwischen dem zu bewertenden Sprachsignal mit einem Referenzsprachsignal Unterschiede zwischen beiden Signalabschnitten in der gewählten Darstellungsebene in den zu berechnenden Qualitätskennwert einfließen, die nicht oder kaum zu einer - auch im auditiven Test wahrnehmbaren - qualitativen Beeinträchtigung führen.A disadvantage of the methods known today in such cases is that a comparison differences between the speech signal to be evaluated with a reference speech signal between the two signal sections in the selected display level in the calculative quality parameter that does not or hardly at all - also in audible test perceptible - lead to qualitative impairment.

    Im Rahmen der hier betrachteten Sprachübertragung in Telefonanwendungen tragen Frequenzbandbegrenzungen und spektrale Verformungen des zu bewertenden Sprachsignals (z.B. hervorgerufen durch Filtereigenschaften des Telefongerätes oder des Übertragungskanals) nur begrenzt zu einer empfundenen qualitativen Beeinträchtigung bei.Wear in the context of the voice transmission considered here in telephone applications Frequency band limits and spectral deformations of the speech signal to be evaluated (e.g. caused by filter properties of the telephone device or the Transmission channel) contributes only to a limited extent to a perceived qualitative impairment.

    Um diese Mängel teilweise zu vermeiden, wird in einem anderen Ansatz versucht, die linearen Verzerrungen (Frequenzgang) durch ein Korrekturfilter bzw. eine Leistungsübertragungsfunktion zu kompensieren (veröffentlicht in: "A new approach to objective quality-measures based on attribute-matching", Halka, U.; Heute, U., Speech communication, 11(1992)1, S. 15-30). Die Anwendung dieses Verfahrens ist jedoch bei nichtlinearer und zeitinvarianter Übertragung nachteilig, da die so berechnete Kompensationsfunktion nicht mehr ausschließlich die spektralen Verformungen des zu bewertenden Signals beschreibt.
    Verschiebungen spektraler Kurzzeit-Maxima ("Formantverschiebungen") im zu testenden Signal gegenüber dem Referenzsprachsignal, z.B. verursacht durch Codiersysteme mit niedriger Bitrate, führen bei bekannten Verfahren zu großen Unterschieden in den spektralen Intensitätsabbildungen und gehen damit stark in den berechneten Qualitätswert ein. Untersuchungen haben ergeben, daß in einer auditiven Sprachqualitätsuntersuchung diese Verschiebungen spektraler Kurzzeit-Maxima jedoch nur begrenzten Einfluß auf das Qualitätsurteil haben.    In order to partially avoid these deficiencies, another approach attempts to compensate for the linear distortions (frequency response) by means of a correction filter or a power transmission function (published in: "A new approach to objective quality-measures based on attribute-matching", Halka , U .; Today, U., Speech communication, 11 (1992) 1, pp. 15-30). However, the use of this method is disadvantageous in the case of nonlinear and time-invariant transmission, since the compensation function calculated in this way no longer exclusively describes the spectral deformations of the signal to be evaluated.
    Shifts in short-term spectral maxima ("formant shifts") in the signal to be tested compared to the reference speech signal, for example caused by coding systems with a low bit rate, lead to large differences in the spectral intensity maps in known methods and are therefore strongly incorporated into the calculated quality value. Studies have shown that, in an auditory speech quality examination, these shifts in short-term spectral maxima have only a limited influence on the quality judgment.

    Aufgabetask

    Die Erfindung stellt sich die Aufgabe, den Einfluß von spektralen Begrenzungen und Verformungen des zu bewertenden Sprachsignals sowie von Verschiebungen spektraler Kurzzeit-Maxima vor dem Vergleich der spektralen Eigenschaften eines zu testenden Signals mit einem Referenzsprachsignal und der Berechnung eines Qualitätswertes in instrumentellen Verfahren zu reduzieren.The invention has as its object the influence of spectral limits and Deformations of the speech signal to be assessed and of spectral shifts Short-term maxima before comparing the spectral properties of a test item Signal with a reference speech signal and the calculation of a quality value in to reduce instrumental procedures.

    Lösungsolution

    Die obengenannte Aufgabe wird durch ein Verfahren gemäß Patentanspruch 1 gelöst.The above object is accomplished by a method according to Claim 1 solved.

    Im Gegensatz zu bekannten Ansätzen wird in der hier beschriebenen Erfindung eine spektrale Wichtungsfunktion generiert, die auf mittleren spektralen Einhüllenden, z.B. der mittleren spektralen Leistungsdichte, von zu bewertendem Sprachsignal und Referenzsprachsignal beruht. Dies ermöglicht den Einsatz des Verfahrens ebenfalls bei nichtlinearer und zeitvarianter Übertragung.In contrast to known approaches, one is used in the invention described here generated spectral weighting function based on medium spectral envelopes, e.g. the average spectral power density, of the speech signal to be evaluated and Reference speech signal based. This also enables the use of the method non-linear and time-variant transmission.

    Die spektrale Wichtungsfunktion wird aus den Quotienten der Stützwerte der mittleren spektralen Leistungsdichte des zu bewertenden Signals PhiY(f) und der des Eingangssignals des Übertragungssystems PhiX(f) derart berechnet, daß die Wichtungsfunktion über WT(f) = a(f)·(PhiY(f) / PhiX(f)) zu beschreiben ist. Die Bewertungsfunktion a(f) kann die Wichtungsfunktion WT(f) an über den Wirkungsbereich unterschiedlich gewichten, sie ist im einfachsten Falle konstant 1.The spectral weighting function is calculated from the quotients of the base values of the mean spectral power density of the signal Phi Y (f) to be evaluated and that of the input signal of the transmission system Phi X (f) in such a way that the weighting function over W T (f) = a (f) · (Phi Y (f) / Phi X (F)) is to be described. The evaluation function a (f) can weight the weighting function W T (f) differently over the effective range, in the simplest case it is constant 1.

    Die derart berechnete spektrale Wichtungsfunktion WT(f) nähert die mittleren spektralen Einhüllenden von zu bewertenden Sprachsignal und Referenzsprachsignal einander an, so daß Unterschiede der beiden spektralen Einhüllenden nur noch vermindert in den berechneten Qualitätswert einfließen. The spectral weighting function W T (f) calculated in this way approximates the mean spectral envelopes of the speech signal and the reference speech signal to be evaluated, so that differences between the two spectral envelopes only have a reduced effect on the calculated quality value.

    Die spektrale Wichtungsfunktion WT(f) kann zum einen auf das Referenzsprachsignal angewendet werden. Dabei wird das Referenzsprachsignal in seiner mittleren spektralen Leistungsdichte dem zu bewertenden Signal angenähert (Fig. 2a).The spectral weighting function W T (f) can be applied to the reference speech signal. The average spectral power density of the reference speech signal is approximated to the signal to be evaluated (FIG. 2a).

    Zum anderen kann die spektrale Wichtungsfunktion invertiert auf das zu bewertende Signal angewendet werden. Dieses wird dadurch entzerrt und, hinsichtlich seiner mittleren spektralen Leistungsdichte, an das Referenzsprachsignal angenähert (Fig. 2b).On the other hand, the spectral weighting function can be inverted to the signal to be evaluated be applied. This is equalized and, with regard to its middle one spectral power density, approximated to the reference speech signal (Fig. 2b).

    Ein weiterer Teil der Erfindung bezieht sich auf die Korrektur von Verschiebungen spektraler Kurzzeit-Maxima, die durch die Übertragungssysteme verursacht werden.Another part of the invention relates to the correction of displacements short-term spectral maxima caused by the transmission systems.

    Die Intensität wird für jeden Zeitabschnitt in Frequenzbändern integriert. Resultat ist eine Reihe von Intensitätswerten für jede spektrale Darstellung eines Signalabschnitts, wobei jeder Einzelwert die Intensität in einem Frequenzband repräsentiert. Die Verschiebungen spektraler Kurzzeit-Maxima können hierbei zu abweichenden berechneten Intensitäten in den Frequenzbändern von Referenzsprachsignal und zu bewertenden Sprachsignal führen.The intensity is integrated in frequency bands for each time period. The result is one Series of intensity values for each spectral representation of a signal section, where each individual value represents the intensity in a frequency band. The shifts Short-term spectral maxima can lead to deviating calculated intensities in the frequency bands of the reference speech signal and the speech signal to be evaluated.

    Diese Abweichungen in den spektralen Intensitätsabbildungen - verursacht Verschiebungen spektraler Kurzzeit-Maxima -können durch eine variable Anordnung der Frequenzbänder auf der Frequenzachse reduziert werden. Im Gegensatz zu den konstanten Bandgrenzen bei bekannten Verfahren werden die Bandgrenzen auf der Frequenzachse verschoben. Die Zahl der Frequenzbänder und deren Index bleibt aber konstant. In einer Optimierungsschleife werden dann diejenigen Bandgrenzen akzeptiert, bei denen die beiden entstehenden spektralen Abbildungen von zu bewertenden Sprachsignal und Referenzsprachsignal maximale Ähnlichkeit aufweisen bzw. deren Abstand minimal ist. Diese Optimierung wird für alle Bänder in allen betrachteten Zeitsegmenten durchgeführt.These deviations in the spectral intensity images - causes shifts spectral short-term maxima - can by variable arrangement of the frequency bands be reduced on the frequency axis. In contrast to the constant band limits at In known methods, the band limits are shifted on the frequency axis. The number the frequency bands and their index remain constant. In an optimization loop the band limits at which the two arise are then accepted spectral images of the speech signal to be evaluated and the reference speech signal have maximum similarity or their distance is minimal. This optimization will carried out for all bands in all considered time segments.

    Der Einsatz variabler Bandgrenzen zur Berechnung der spektralen Intensitätsabbildung ist nicht nur auf das Signal, in dem auch die beschriebene spektrale Wichtungsfunktion WT(f) zum Einsatz kommt, beschränkt, sondern kann auch auf das jeweils andere Signal und sogar auf beide Signale angewendet werden. (vgl. Fig. 2a und 2b). The use of variable band limits for calculating the spectral intensity mapping is not only limited to the signal in which the described spectral weighting function W T (f) is also used, but can also be applied to the other signal and even to both signals. (see FIGS. 2a and 2b).

    Ausführungsbeispiel:Embodiment:

    Ein spezielles Ausführungsbeispiel zeigt eine Realisierung gemäß Fig. 3, die als TOSQA (Telecommunication Objective Speech Quality Assessment) bezeichnet wird. Hierbei erfolgt eine erweiterte Vorverarbeitung des Referenzsprachsignals.A special embodiment shows an implementation according to FIG. 3, which is called TOSQA (Telecommunication Objective Speech Quality Assessment). This is done advanced preprocessing of the reference speech signal.

    In Spezifikation der allgemeinen Realisierungen nach Fig. 2a und 2b werden hier Sprachpausen mittels eines Sprachpausenerkenners erkannt und gehen nicht in das Qualitätsmaß ein. Ebenfalls erfolgt eine Filterung von Referenzsprachsignal und zu bewertendem Sprachsignal mit einem Bandpaß 300...3400 Hz sowie eine Filterung auf den Frequenzgang eines Telefonhandapparates. Die Integration der spektralen Leistungsdichte erfolgt in Frequenzgruppen, die die Basis für die Berechnung der spezifischen Lautheit darstellen.In specification of the general realizations according to FIGS. 2a and 2b are here Speech pauses are recognized by a speech pause recognizer and do not go into that Quality measure. The reference speech signal and is also filtered evaluating speech signal with a bandpass 300 ... 3400 Hz and a filtering on the Frequency response of a telephone handset. The integration of the spectral power density takes place in frequency groups, which are the basis for the calculation of specific loudness represent.

    Die Integration in Frequenzgruppen erfolgt jedoch nicht in festen Frequenzgruppengrenzen, sondern mit den in dieser Erfindung beschriebenen variablen Frequenzgruppengrenzen. Die berechneten Signalleistungen in den so modifizierten Frequenzgruppen bilden die Basis für die Intensitätsberechnung. Hier wurde auf ein Modell zur Berechnung der spezifischen Lautheit nach Zwicker, einer gehörrichtigen Intensitätsabbildung, zurückgegriffen (veröffentlicht in Zwicker, E.: "Psychoakustik", Berlin: Springer-Verlag, 1982).However, the integration into frequency groups does not take place in fixed frequency group limits, but with the variable frequency group limits described in this invention. The calculated signal powers in the frequency groups modified in this way form the basis for the intensity calculation. Here, a model was used to calculate the specific loudness according to Zwicker, an aurally accurate intensity mapping (published in Zwicker, E .: "Psychoacoustics", Berlin: Springer-Verlag, 1982).

    Die berechneten Lautheitsmuster werden in Ergänzung des allgemeinen Ansatzes noch durch eine Fehlerbewertungsfunktion ergänzt. Der berechnete Qualitätswert wird über einen Mittelwert der Korrelationskoeffizienten der spezifischen Lautheiten für jedes betrachtete kurze Zeitsegment über die Zahl der ausgewerteten Sprachsegmente gebildet.The calculated loudness patterns are in addition to the general approach supplemented by an error evaluation function. The calculated quality value is over a Average of the correlation coefficients of the specific loudness for each considered short time segment over the number of evaluated language segments formed.

    Claims (6)

    1. Method for instrumental speech quality determination in which characteristic values for determining the speech quality are computed by comparing spectral short-time properties of a speech signal to be evaluated with a reference speech
      signal, characterized in that
      prior to comparison of the properties of the speech signals, differences in mean spectral envelope curves are reduced in that first a spectral weighting function is computed therefrom, said spectral weighting function being used to weight the spectral short-time properties of the speech signals in all time segments under consideration, with the result that the differences in the mean spectral envelope curves are thereby included only to a limited extent in the quality characteristic value to be computed, and
      in that, for computing the signal intensity, the limits of the frequency bands used are made variable, with the result that, for each signal portion under consideration, the computed intensities of reference speech signal and signal to be evaluated have differences as small as possible with respect to each other in all evaluated frequency bands.
    2. Method according to claim 1, characterized in that first the mean spectral envelope curves of speech signal to be evaluated and reference speech signal are computed in the form of a mean power density spectrum and a spectral weighting function WT(f) is computed from the quotient of both spectra, said spectral weighting function WT(f) being used to weight the short-time power density spectra of the reference speech signal prior to the computation of a quality characteristic value.
    3. Method according to claims 1 and 2, characterized in that the weighting function WT(f) to be computed is computed only from partial regions of the computed mean spectral envelope curves of speech signal to be evaluated and reference speech signal and, consequently, the differences in mean spectral envelope curves between both signals are reduced only in spectral partial regions.
    4. Method according to claims 1 to 3, characterized in that, prior to computation of the quality characteristic values, the signal intensity for each evaluated short time portion is integrated in frequency groups, the limits of the frequency groups being variable on the frequency axis, but the width of the frequency groups remaining constant on the critical band rate scale, and in that from the signal intensities in the frequency groups a computation is made of the specific loudness, use being made of the limits of the frequency groups in which the computed differences in the specific loudness between the signal to be evaluated and the reference speech signal have the smallest difference in the respective band and time segment under consideration.
    5. Method according to claims 1 to 4, characterized in that the quality characteristic value is computed from the similarity of the spectral representations in each time portion under consideration, the similarity representing a correlation coefficient averaged over all time portions under consideration between the spectral representation of the speech signal to be evaluated and the spectral representation of the reference speech signal in the respective time segment.
    6. Method according to claim 5, characterized in that the correlation coefficient between the spectral representation of the speech signal to be evaluated and the spectral representation of the reference speech signal in the respective time segment is computed only from a partial region of the spectral representation, i.e. not all the computed spectral values are taken into consideration for the computation of the quality characteristic value.
    EP99942871A 1998-08-27 1999-08-14 Method for objective voice quality evaluation Expired - Lifetime EP1048025B1 (en)

    Applications Claiming Priority (3)

    Application Number Priority Date Filing Date Title
    DE19840548 1998-08-27
    DE19840548A DE19840548C2 (en) 1998-08-27 1998-08-27 Procedures for instrumental language quality determination
    PCT/EP1999/005972 WO2000013173A1 (en) 1998-08-27 1999-08-14 Method for instrumental voice quality evaluation

    Publications (2)

    Publication Number Publication Date
    EP1048025A1 EP1048025A1 (en) 2000-11-02
    EP1048025B1 true EP1048025B1 (en) 2003-11-05

    Family

    ID=7879918

    Family Applications (1)

    Application Number Title Priority Date Filing Date
    EP99942871A Expired - Lifetime EP1048025B1 (en) 1998-08-27 1999-08-14 Method for objective voice quality evaluation

    Country Status (6)

    Country Link
    US (1) US7013266B1 (en)
    EP (1) EP1048025B1 (en)
    AT (1) ATE253765T1 (en)
    CA (1) CA2305652A1 (en)
    DE (2) DE19840548C2 (en)
    WO (1) WO2000013173A1 (en)

    Families Citing this family (12)

    * Cited by examiner, † Cited by third party
    Publication number Priority date Publication date Assignee Title
    WO2001065543A1 (en) * 2000-02-29 2001-09-07 Telefonaktiebolaget Lm Ericsson (Publ) Compensation for linear filtering using frequency weighting factors
    EP1241663A1 (en) * 2001-03-13 2002-09-18 Koninklijke KPN N.V. Method and device for determining the quality of speech signal
    EP1292036B1 (en) * 2001-08-23 2012-08-01 Nippon Telegraph And Telephone Corporation Digital signal decoding methods and apparatuses
    DE10142846A1 (en) * 2001-08-29 2003-03-20 Deutsche Telekom Ag Procedure for the correction of measured speech quality values
    DE10150519B4 (en) 2001-10-12 2014-01-09 Hewlett-Packard Development Co., L.P. Method and arrangement for speech processing
    EP1492084B1 (en) * 2003-06-25 2006-05-17 Psytechnics Ltd Binaural quality assessment apparatus and method
    US7305341B2 (en) 2003-06-25 2007-12-04 Lucent Technologies Inc. Method of reflecting time/language distortion in objective speech quality assessment
    WO2006033570A1 (en) * 2004-09-20 2006-03-30 Nederlandse Organisatie Voor Toegepast- Natuurwetenschappelijk Onderzoek Tno Frequency compensation for perceptual speech analysis
    EP2249333B1 (en) * 2009-05-06 2014-08-27 Nuance Communications, Inc. Method and apparatus for estimating a fundamental frequency of a speech signal
    EP2388779B1 (en) * 2010-05-21 2013-02-20 SwissQual License AG Method for estimating speech quality
    WO2013142695A1 (en) * 2012-03-23 2013-09-26 Dolby Laboratories Licensing Corporation Method and system for bias corrected speech level determination
    CN112233693B (en) * 2020-10-14 2023-12-01 腾讯音乐娱乐科技（深圳）有限公司 Sound quality evaluation method, device and equipment

    Family Cites Families (6)

    * Cited by examiner, † Cited by third party
    Publication number Priority date Publication date Assignee Title
    DE3708002A1 (en) * 1987-03-12 1988-09-22 Telefonbau & Normalzeit Gmbh Measuring method for assessing the quality of speech coders and/or transmission routes
    US4860360A (en) * 1987-04-06 1989-08-22 Gte Laboratories Incorporated Method of evaluating speech
    GB9213459D0 (en) 1992-06-24 1992-08-05 British Telecomm Characterisation of communications systems using a speech-like test stimulus
    SE517836C2 (en) * 1995-02-14 2002-07-23 Telia Ab Method and apparatus for determining speech quality
    NL9500512A (en) * 1995-03-15 1996-10-01 Nederland Ptt Apparatus for determining the quality of an output signal to be generated by a signal processing circuit, and a method for determining the quality of an output signal to be generated by a signal processing circuit.
    EP0809236B1 (en) * 1996-05-21 2001-08-29 Koninklijke KPN N.V. Device for determining the quality of an output signal to be generated by a signal processing circuit, and also method

    Also Published As

    Publication number Publication date
    DE19840548C2 (en) 2001-02-15
    US7013266B1 (en) 2006-03-14
    DE59907623D1 (en) 2003-12-11
    EP1048025A1 (en) 2000-11-02
    CA2305652A1 (en) 2000-03-09
    ATE253765T1 (en) 2003-11-15
    DE19840548A1 (en) 2000-03-02
    WO2000013173A1 (en) 2000-03-09

    Similar Documents

    Publication Publication Date Title
    DE112017004548B4 (en) Method and apparatus for robust noise estimation for speech enhancement in variable noise conditions
    DE60009206T2 (en) Noise suppression by means of spectral subtraction
    DE10041512B4 (en) Method and device for artificially expanding the bandwidth of speech signals
    DE19952538C2 (en) Automatic gain control in a speech recognition system
    DE60020865T2 (en) System, method and computer program for a telephone emotion detector with feedback to an operator
    DE60031432T2 (en) SYSTEM, METHOD, AND MANUFACTURED SUBJECT FOR DETECTING EMOTIONS IN LANGUAGE SIGNALS BY STATISTICAL ANALYSIS OF LANGUAGE SIGNAL PARAMETERS
    EP1048025B1 (en) Method for objective voice quality evaluation
    EP0938831B1 (en) Hearing-adapted quality assessment of audio signals
    DE60122751T2 (en) METHOD AND DEVICE FOR OBJECTIVE EVALUATION OF LANGUAGE QUALITY WITHOUT REFERENCE SIGNAL
    DE69730721T2 (en) METHOD AND DEVICES FOR NOISE CONDITIONING OF SIGNALS WHICH REPRESENT AUDIO INFORMATION IN COMPRESSED AND DIGITIZED FORM
    EP0980064A1 (en) Method for carrying an automatic judgement of the transmission quality of audio signals
    KR19990028694A (en) Method and device for evaluating the property of speech transmission signal
    DE602004010634T2 (en) METHOD AND SYSTEM FOR LANGUAGE QUALITY FORECASTING AN AUDIO TRANSMISSION SYSTEM
    DE69918635T2 (en) Apparatus and method for speech processing
    DE3043516C2 (en) Method and device for speech recognition
    DE602004008666T2 (en) Tracking vocal tract resonances using a nonlinear predictor
    DE10254612A1 (en) Method for determining specifically relevant acoustic characteristics of sound signals for the analysis of unknown sound signals from a sound generation
    EP0772764B1 (en) Process and device for determining the tonality of an audio signal
    EP3291234A1 (en) Method for evaluation of a quality of the voice usage of a speaker
    DE69922769T2 (en) Apparatus and method for speech processing
    EP1382034B1 (en) Method for determining intensity parameters of background noise in speech pauses of voice signals
    DE60110541T2 (en) Method for speech recognition with noise-dependent normalization of the variance
    DE60305306T2 (en) Apparatus and method for binaural quality assessment
    EP0535425A2 (en) Method for amplifying an acoustic signal for the hard of hearing and device for carrying out the method
    DE602004011292T2 (en) Device for speech detection

    Legal Events

    Date Code Title Description
    PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

    Free format text: ORIGINAL CODE: 0009012

    AK Designated contracting states

    Kind code of ref document: A1

    Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT SE

    17P Request for examination filed

    Effective date: 20000911

    RTI1 Title (correction)

    Free format text: METHOD FOR OBJECTIVE VOICE QUALITY EVALUATION

    RTI1 Title (correction)

    Free format text: METHOD FOR OBJECTIVE VOICE QUALITY EVALUATION

    GRAH Despatch of communication of intention to grant a patent

    Free format text: ORIGINAL CODE: EPIDOS IGRA

    GRAS Grant fee paid

    Free format text: ORIGINAL CODE: EPIDOSNIGR3

    GRAA (expected) grant

    Free format text: ORIGINAL CODE: 0009210

    AK Designated contracting states

    Kind code of ref document: B1

    Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT SE

    PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

    Ref country code: NL

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

    Effective date: 20031105

    Ref country code: IT

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT;WARNING: LAPSES OF ITALIAN PATENTS WITH EFFECTIVE DATE BEFORE 2007 MAY HAVE OCCURRED AT ANY TIME BEFORE 2007. THE CORRECT EFFECTIVE DATE MAY BE DIFFERENT FROM THE ONE RECORDED.

    Effective date: 20031105

    Ref country code: IE

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

    Effective date: 20031105

    Ref country code: FI

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

    Effective date: 20031105

    Ref country code: CY

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

    Effective date: 20031105

    REG Reference to a national code

    Ref country code: GB

    Ref legal event code: FG4D

    Free format text: NOT ENGLISH

    REG Reference to a national code

    Ref country code: CH

    Ref legal event code: NV

    Representative=s name: ISLER & PEDRAZZINI AG

    Ref country code: CH

    Ref legal event code: EP

    REF Corresponds to:

    Ref document number: 59907623

    Country of ref document: DE

    Date of ref document: 20031211

    Kind code of ref document: P

    REG Reference to a national code

    Ref country code: IE

    Ref legal event code: FG4D

    Free format text: GERMAN

    PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

    Ref country code: SE

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

    Effective date: 20040205

    Ref country code: GR

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

    Effective date: 20040205

    Ref country code: DK

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

    Effective date: 20040205

    PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

    Ref country code: ES

    Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

    Effective date: 20040216

    GBT Gb: translation of ep patent filed (gb section 77(6)(a)/1977)

    Effective date: 20040211

    NLV1 Nl: lapsed or annulled due to failure to fulfill the requirements of art. 29p and 29m of the patents act
    REG Reference to a national code

    Ref country code: IE

    Ref legal event code: FD4D

    ET Fr: translation filed
    PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

    Ref country code: LU

    Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

    Effective date: 20040814

    PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

    Ref country code: MC

    Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

    Effective date: 20040831

    Ref country code: BE

    Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

    Effective date: 20040831

    PLBE No opposition filed within time limit

    Free format text: ORIGINAL CODE: 0009261

    STAA Information on the status of an ep patent application or granted ep patent

    Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

    26N No opposition filed

    Effective date: 20040806

    BERE Be: lapsed

    Owner name: DEUTSCHE *TELEKOM A.G.

    Effective date: 20040831

    REG Reference to a national code

    Ref country code: CH

    Ref legal event code: PCAR

    Free format text: ISLER & PEDRAZZINI AG;POSTFACH 1772;8027 ZUERICH (CH)

    BERE Be: lapsed

    Owner name: DEUTSCHE *TELEKOM A.G.

    Effective date: 20040831

    PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

    Ref country code: PT

    Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

    Effective date: 20040405

    REG Reference to a national code

    Ref country code: FR

    Ref legal event code: PLFP

    Year of fee payment: 18

    REG Reference to a national code

    Ref country code: FR

    Ref legal event code: PLFP

    Year of fee payment: 19

    REG Reference to a national code

    Ref country code: FR

    Ref legal event code: PLFP

    Year of fee payment: 20

    PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

    Ref country code: DE

    Payment date: 20180827

    Year of fee payment: 20

    Ref country code: FR

    Payment date: 20180824

    Year of fee payment: 20

    PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

    Ref country code: AT

    Payment date: 20180821

    Year of fee payment: 20

    Ref country code: GB

    Payment date: 20180828

    Year of fee payment: 20

    Ref country code: CH

    Payment date: 20180827

    Year of fee payment: 20

    REG Reference to a national code

    Ref country code: DE

    Ref legal event code: R071

    Ref document number: 59907623

    Country of ref document: DE

    REG Reference to a national code

    Ref country code: CH

    Ref legal event code: PL

    REG Reference to a national code

    Ref country code: GB

    Ref legal event code: PE20

    Expiry date: 20190813

    REG Reference to a national code

    Ref country code: AT

    Ref legal event code: MK07

    Ref document number: 253765

    Country of ref document: AT

    Kind code of ref document: T

    Effective date: 20190814

    PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

    Ref country code: GB

    Free format text: LAPSE BECAUSE OF EXPIRATION OF PROTECTION

    Effective date: 20190813