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
  • G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
  • G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
  • G10L21/0208—Noise filtering

Definitions

• the present invention is generally in the fields of speech signal processing and psychoacoustics. More specifically, the invention relates to a method and a device for objective evaluation of the annoyance due to noise in audio signals.
• a noise reduction function also known as a noise canceling or denoising function
• a noise reduction function is intended to reduce the background noise level in a voice communication, or having at least one component voice. It has a specific interest when one of the interlocutors of this communication is immersed in a noisy environment that greatly impairs the intelligibility of his voice.
• the noise reduction algorithms are based on a continuous estimation of the background noise level from the incident signal and a speech activity detection to distinguish the noise periods only from those with the useful speech signal. A filtering of the incident speech signal corresponding to the noisy speech signal is then performed to reduce the noise contribution determined from the noise estimate.
• the invention will be used to evaluate noise annoyance at the output of communication equipment implementing a noise reduction function, the invention also applies to noisy signals. not treated by such a function.
• the case of use of the invention on any noisy audio signal is therefore a particular case of the more general case of use of the invention on an audio signal processed by a noise reduction function.
• the present invention aims to overcome the disadvantages of the prior art by providing a method and an objective computing device of a score equivalent to the subjective score as indicated in the document "ITU-T Recommendation P.835", characterizing the annoyance due to the presence of noise in an audio signal.
• the method according to the invention varies according to whether the invention is used on any noisy audio signal or on an audio signal processed by a noise reduction function, in particular in the parameters for calculating the objective score according to the invention.
• two embodiments that can also be considered as two distinct processes are presented.
• the second embodiment applying to any noisy audio signal, and more general than the first embodiment, is easily deduced therefrom.
• the invention proposes a method of calculating an objective note of the annoyance due to noise in an audio signal processed by a noise reduction function, said method comprising a preliminary step of obtaining an audio signal.
• predefined test device comprising a useful signal devoid of noise, a noisy signal obtained by adding a predefined noise signal to said test signal, and a processed signal obtained by applying the noise reduction function to said noisy signal said method being characterized in that it includes a step of measuring loudness of frames of said noisy signal and said processed signal, and measures of frame pitch coefficients of said processed signal.
• this method according to the invention comprises the steps of;
• the step of calculating average loudness densities and tonality coefficients is followed by a step averaging & - &", pam ie, Sr Pamie, 5V bn, u ar _ and brown said average loudness densities and said tone coefficients over all the relevant frames of the corresponding signals, and the objective noise noise score is calculated according to the following equation: s NOB -] jT ⁇ factor (i) + o & ,
• SY_ word factor (3) Type_offset _ word) - Sr (m _ speech)), the operator
• the invention also relates to a method for calculating an objective note of the annoyance due to noise in an audio signal, said method comprising a preliminary step of obtaining a predefined test audio signal containing a useful signal devoid of noise, and a noisy signal obtained by adding a predefined noise signal to said test signal, said method being characterized by including a loudness measurement and frame tone coefficient measurement of said noisy signal.
• This method has the same advantages as the previous method, but applies to any noisy audio signal.
• this method according to the invention comprises the steps of:
• the step of calculating mean loudness densities and tone coefficients is followed by a calculation step average s M, S x b ⁇ speech, MS ⁇ brmt and axi, bmn desd SIU densities average loudness of said tone coefficients over all the relevant frames of the corresponding signals, and in that said objective score of the noise interference is calculated according to the following equation:
• the coefficients of this linear combination have the advantage of being able to be recalculated if new subjective test data substantially modify the previously established correlation. This makes it possible to improve an objective model fed by the method according to the invention, of calculating the annoyance due to the noise in an audio signal, by a simple reconfiguration of the parameters of the method.
• the step of calculating loudness densities and tone coefficients is preceded by a voice activity detection step sot the test signal, so as to determine if a current frame of the noisy signal, and the signa! treated in the case of the first method, is a "m_noise" frame containing only noise, or a "m_parole” frame containing speech, called a useful signal frame.
• This voice activity detection step makes it possible to very simply separate the different types of frames of the noisy signal, and of the signal processed in the case of the first method, by the use of the test signal.
• the step of calculating the objective score is followed by a step of calculating an objective score on the MOS scale of the annoyance due to the noise, calculated according to the following equation:
• NOB _ MOS ⁇ t (NOB) '- 1 ,
• the step of calculating loudness densities and tone coefficients, calculating the average loudness density Su (m) of a frame of any index m, a given audio signal u comprises the following steps:
• windowing for example of the Hanning type, of the frame of index m and obtaining a windowed frame u_w [m],
• the calculation of the tone coefficient ⁇ (m) of a frame of any index m of a given audio signal u comprises the following steps: windowing, for example of the Hanning type, of the frame of index m and obtaining a windowed frame u_w [m], applying a fast Fourier transform to the windowed frame u w w [m] and obtaining a corresponding frame U (m, f) in the frequency domain, calculating the power spectral density ⁇ u (m, f) of the frame U (m, f), calculation of the tone coefficient ⁇ (m) according to the following equation:
• f represents the frequency index of the power spectral density
• N denotes the size of the fast Fourier transform
• the invention also relates to a test equipment for evaluating an objective note of the annoyance due to noise in an audio signal, characterized in that it comprises means adapted to implement one or the other of the methods according to the invention.
• the test equipment includes computer means and a computer program, said program comprising instructions adapted to implement one or the other of said methods, when it is executed by said computer means. .
• the invention also relates to a computer program on an information carrier " comprising instructions adapted to the implementation of one or the other of the methods according to the invention, when the program is loaded and executed in a computer system.
• FIG. 1 represents a test environment intended to calculate an objective score of the annoyance due to the noise in an audio signal processed by a noise reduction function, according to a first embodiment of the invention
• FIG. 2 is a flowchart illustrating a method for calculating an objective note of the annoyance due to noise in an audio signal processed by a noise reduction function according to a first embodiment of the method according to the invention
• FIG. 3 is a flowchart illustrating a method for calculating an objective note of the annoyance due to noise in an audio signal according to a second embodiment of the method according to the invention
• FIG. 4 is a flowchart illustrating the method of calculating the mean loudness density and the tone coefficient of an audio signal frame according to the invention.
• the first being applied to an audio signal processed by a noise reduction function
• the second being applied to any noisy audio signal.
• the principle of the method according to the invention is the same in these two embodiments, in particular the calculation method is exactly the same, but in the second embodiment ie signa! audio processed by a noise reduction function is taken equal to the signa! noisy.
• the second embodiment can indeed be considered as a special case of the first embodiment, with an inhibited noise reduction function.
• the annoyance due to the presence of noise in an audio signal processed by a function of noise reduction is objectively evaluated in a test environment shown in FIG. 1.
• Such a test environment comprises a source of SSA audio signals delivering a test audio signal x (n) containing only the wanted signal, and that is to say, without noise, for example a speech signal, and a noise source SB delivering a predefined noise signal.
• this predefined noise signal is added to the selected test signal x (n), as represented by the AD addition operator.
• the audio signal resulting from this addition of noise to the test signal x (n) is denoted xb (n) and is designated by the expression "noisy signal”.
• the noisy signal xb (n) then constitutes the input signal of a module
• MRB noise reduction implementing a noise reduction function outputting an audio signal y (n) designated by the expression "processed signal".
• the processed signal y (n) is therefore an audio signal containing useful signal and residual noise.
• the processed signal y (n) is then delivered to a test equipment EQT implementing a method of objective evaluation of the annoyance due to the noise in the processed signal, according to the invention.
• the method according to the invention is implemented in the EQT test equipment in the form of a computer program.
• the EQT test equipment optionally comprises electronic hardware to implement the method according to the invention.
• the test equipment EQT receives as input the test signal x (n) and the noisy signal xb (n). The EQT test equipment outputs an evaluation result
• the aforementioned audio signals x (n), xb (n) and y (n) are signals sampled in a digital format, n designating a sample any. These signals are for example supposed to be sampled at the sampling frequency of 8 kHz (kilo Hertz).
• the test signal x (n) is a speech signal devoid of noise.
• the noisy signal xb (n) then represents the initial speech signal x (n) degraded by a noisy environment (background or ambient noise), and the signal y (n) represents the signal xb (n) after noise reduction.
• the signal x (n) is generated in an anechoic chamber.
• the signal x (n) can also be generated in a "quiet" room having an "average" reverberation time of less than 0.5 seconds.
• the noisy signal xb (n) is obtained by adding a predetermined contribution of noise to the signal x (n).
• the signal y (n) is obtained either at the output of a noise reduction algorithm implanted on a personal computer, or at the output of a noise reduction network equipment and in the latter case, the signal y (n) is taken at the level of a PCM encoder (pulse modulation and coding).
• the method for calculating the objective note NOB_MOS of the annoyance due to the noise in the processed signal y (n) according to the invention is represented in the form of an algorithm comprising steps a1 to a7.
• ai the signals x (n), xb (n) and y (n) are respectively divided into successive time windows called frames.
• Each signal frame, noted m, contains a predetermined number of samples of the signal, step a thus consists of a change of rate of each of these signals.
• the signals x (n), xb (n) and y ⁇ n) passed in frame rate respectively produce the signals x [m], xb [m], and y [m].
• a second step a2 voice activity detection (DAV) is performed on the signal x [m] so as to determine whether each respective current frame of index m has signals xb [m] and y [m]. is a frame containing only noise, denoted "m noise”, or a frame containing ia speech, that is to say the useful signal, and noted “m_parole”. This determination is made by comparing the signals xb [m] and y [m] with the test signal x [m] devoid of noise.
• DAV voice activity detection
• Each silence frame of x [m] corresponds in fact to a noise frame for the signals xb [m] and y [m], while each speech frame of x [m] corresponds to a speech frame for the signals xb [m] and y [m].
• step a2 As represented in FIG. 2, at the output of step a2, three types of frames are selected from the signals x [m], xb [m] and y [m]:
• a third step a3 loudness measurements are made on at least sets of y [m_noise], y [m_parole], xb [m_parole] frames from the previous step a2, and at least one set of frames of the signal y [m] at the output of step ai. For example, if 8 seconds of sampled test signal at 8 kHz is used, it will be possible to work on 250 fields y [m] of 256 samples of signal y (n). In addition, the tone coefficients of at least one set of y [m_noise] frames are measured.
• this step one calculates the loudness densities average ⁇ S ⁇ b (word m_), ⁇ S ⁇ ⁇ ⁇ m_ speech), ⁇ S ⁇ ⁇ m), and £ y (m _bruit) of respectively each of the frames xb [m_parole], y [m_parole], y [m] and y [rn_noise] sets of frames considered.
• the tone coefficients a > - (m noise) of each of the frames y [m_noise] of the considered set of frames y [m] are calculated.
• the calculation of an average loudness density Su (m) and a tone coefficient ⁇ (m) of a frame of any index m of a given audio signal u will be detailed later in connection with FIG. 4.
• NOB is obtained by linear combination of the five factors calculated in step a5, according to the following equation:
• the coefficients ⁇ i to ⁇ % are predefined weighting coefficients. These coefficients were determined in order to obtain a maximum correlation between the subjective data from a subjective test database, and the objective scores NOB calculated by this linear combination using the test signals, noisy and processed x [m ], xb [m] and y [m] used in these same subjective tests.
• the subjective test database is, for example, a database of scores obtained with groups of listeners in accordance with the "Recommendation UiT-T P.835 TM, in which these notes are called background noise notes. It should be noted that the obtaining of the weighting coefficients by the use of a database of subjective tests is not essential for each step of calculating an objective score NOB. Indeed, these coefficients must be obtained prior to the first use of the process, and may be the same for all uses of the process. These coefficients are nevertheless likely to evolve when new subjective data come to feed the database of subjective tests used.
• an objective note NOB_MOS of the annoyance due to the noise in the processed signal y (n) on the MOS scale is calculated using for example a polynomial function of order 3, according to the following equation:
• NOB _ MOS ⁇ X 1 (NOB) w ,
• the annoyance due to the presence of noise in any noisy audio signal is evaluated objectively.
• the same test environment is used as in Figure 1, but by removing the MRB noise reduction module.
• the audio signal source SSA delivers a test audio signal x (n) containing only the wanted signal, to which is added a predefined noise signal generated by the noise source SB, to obtain at the output of the addition operator AD a noisy signal xb (n).
• the test signal x (n) and the noisy signal xb (n) are then directly sent to the input of the test equipment EQT implementing a method of objective evaluation of the annoyance due to the noise in the noisy signal.
• the signals x (n) and xb (n) are assumed to be sampled at the sampling frequency 8 kHz.
• the test equipment EQT outputs an evaluation result RES, which is an objective note NOB_MOS of the annoyance due to the presence of noise in the noisy signal xb (n).
• the method for calculating the objective noise-noise rating NOB-MOS in the noisy signal xb (n) according to the invention is represented in the form of an algorithm comprising steps b1 to b7.
• a first step b1 the signals x (n) and xb (n) are split into frames x [m] and xb [m] of time index m.
• a voice activity detection is performed on the signal x [m] so as to determine whether each current frame of index m of the noisy signal xb [m] is a frame containing only noise, denoted " m_noise ", or a frame also containing speech, denoted” m_parole ".
• Two types of frames are thus selected from the signals x [m] and xb [m] at the output of step b2: the speech frames of the noisy signal xb [m], denoted xb [m_parole],
• a third step b3 loudness measurements are made on at least sets of frames xb [m_noise] and xb [m_parole] from the previous step b2, and at least one set of frames of the signal xb [m] in exit from step b1. More tone coefficients * at least one set of frames kB [m_noise] are measured.
• the average output densities Si /. (»I), ⁇ Sxb (m _ speech) and ⁇ S ⁇ b (m _hruit) of respectively each of the frames xb [m], xb [m_parole] are calculated. and xb [m_noise] sets of frames considered.
• the tone coefficients a-Mm noise) of each frames xb [m_noise] of the considered set of frames xb [m_noise] are calculated.
• the average ton coefficients aw (m_noise) previously calculated on the considered set of frames xb [m_noise] is also calculated.
• step b6 the calculation of an intermediate objective score NOB is obtained by linear combination of the four factors calculated in step b5, according to the following equation:
• the coefficients ⁇ i to ⁇ s are predefined weighting coefficients. These coefficients were determined in order to obtain a maximum correlation between the subjective data from a subjective test database, and the objective scores NOB calculated by this linear combination using the test signals and the noisy signals x [m ] and xb [m] used in these same subjective tests.
• obtaining the weighting coefficients by the use of a database of subjective tests is not indispensable at each step of calculating an objective score NOB.
• an objective note NOBJVlOS of the annoyance due to the noise in the noisy signal xb (n) on the MOS scale is calculated using for example a polynomial function of order 3, according to the following equation:
• NOB _ MOS ] T ⁇ i (NOB) '-', i- ⁇ where the coefficients ⁇ i to ⁇ 4 are determined in such a way that the objective score obtained NOB_MOS characterizes the annoyance due to the noise on the MOS scale, c on a scale of 1 to 5.
• the calculation according to the invention of the mean loudness density Su (m) of a frame of any index m of a given audio signal u [m], comprises the steps d to c7 shown in FIG. -after.
• the calculation according to the invention of the tone coefficient ⁇ (m) of a frame of any index m of a given audio signal u [m] comprises the steps d, c2, c3 and c8 shown in FIG. described below.
• the signal u [mj represents any of the signals x [m], xb [m], or y [m] defined above.
• a first step " is applied to ia frame of index m of the signal u [mj windowing, e.g. type windowing Hanning, Hamming or equivalent. We then obtain a windowed frame u_w [m].
• a fast Fourier transform FFT is applied to the windowed frame u_w [m] and a corresponding frame U (m, f) in the frequency domain is accordingly obtained.
• the power spectral density ⁇ y (m, f) of the frame U (m, f) is calculated. Such a calculation is known to those skilled in the art and will not, therefore, be detailed here.
• step c8 is used to calculate the coefficient of tone, then at step c4 for the calculation of the average loudness density Su (m), since for these two signals the two calculations are necessary.
• step c4 is used to calculate the average loudness density Su (m). It should be noted that the calculation of the tone coefficient is independent of the calculation of the mean loudness density Su (m), the two calculations can therefore be carried out in parallel or one after the other.
• step c4 a frequency conversion of the frequency axis at the Barks scale is applied to the power spectral density ⁇ u (m, f) obtained in the previous step, and a spectral density is consequently obtained.
• power, B, j (m, b), on the Barks scale also called Bark's spectrum.
• B, j (m, b) on the Barks scale
• 18 critical bands must be considered. This type of conversion is known to those skilled in the art, the principle of this Hertz / Bark conversion consists in adding all the frequential contributions present in the critical band of the Barks scale.
• E 1J (m, b) obtained previously in loudness densities expressed in sones.
• a calibration of the spectral density spread on the Barks scale, Ey (m, b) is performed by the respective power scaling and loudness scaling factors commonly used in psychoacoustics.
• the size obtained is then converted on the scale of the phones.
• the conversion on the scale of the phones is carried out based on the isosonic curves (Fletcher curves) in accordance with the standard NF ISO 226 "Normal isosonic lines".
• the conversion to sones is effected in accordance with Zwicker's law according to which: f N (phonc) ⁇ 0 x ;
• step c6 there is a number B of loudness density values, S 1 ; (m, b), of the frame of index m for the critical band b, B being the number of critical bands considered in the Barks scale and the index b varying from 1 to B.
• step c7 the average loudness density Su (m) of the frame of index m is calculated from said B loudness density values, according to the following equation:
• the average loudness density Su (m) according to the invention of a frame of index m is therefore the average of the B loudness density values S LI (m, b), of the frame of index m for a critical band b considered.
• step c8 the tone coefficient ⁇ (m) of the frame of index m is calculated according to the following equation:
• the tone coefficient ⁇ of a basic signal is a measure to show whether certain pure frequencies emerge from this signal. It is equivalent to a tonal density. Indeed, the closer the tone coefficient ⁇ is to 0, the more the signal is assimilated to noise, the more the tone coefficient ⁇ is close to 1, the more the signal is component tonal majority. A tone coefficient ⁇ close to 1 attests to the presence of useful signal, or speech signal.

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)
• Quality & Reliability (AREA)
• Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
• Signal Processing Not Specific To The Method Of Recording And Reproducing (AREA)
• Noise Elimination (AREA)

Applications Claiming Priority (2)

Publications (2)

Family

ID=34981381

Family Applications (1)

Country Status (7)

Families Citing this family (9)

Family Cites Families (9)

• 2005
  • 2005-02-18 FR FR0501747A patent/FR2882458A1/fr active Pending
• 2006
  • 2006-02-13 AT AT06709505T patent/ATE438173T1/de not_active IP Right Cessation
  • 2006-02-13 WO PCT/FR2006/050126 patent/WO2006087490A1/fr active Application Filing
  • 2006-02-13 EP EP06709505A patent/EP1849157B1/de not_active Not-in-force
  • 2006-02-13 ES ES06709505T patent/ES2329932T3/es active Active
  • 2006-02-13 US US11/884,573 patent/US20080267425A1/en not_active Abandoned
  • 2006-02-13 DE DE602006008111T patent/DE602006008111D1/de not_active Expired - Fee Related

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events