DETAILED ACTION
Introduction
Applicant's submission filed on 06/15/2026 has been entered. Claims 1-16 are pending in the application and have been examined.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
The response filed on 06/15/2026 has been correspondingly accepted and considered in this Office Action. Claims 1-16 have been examined. New claim 16 has been included and examined in this office action.
Response to Arguments
Applicant's arguments filed 06/15/2026 have been fully considered as follows:
Applicant’s arguments with respect to claim 1 (also representative of claim 14) state that
“Applicant respectfully submits that the combination of Borgstrom and Gallart does not teach or suggest the subject matter recited in amended claim 1. In particular, neither Borgstrom nor Gallart, individually or in combination, teaches or suggests "generating, by a Wiener filter implemented via the CPU, a background noise estimate based on band energies of frames identified as noise frames using both the first band gains and the voice activity detection value," as recited in amended claim 1.”
Applicant’s arguments above with respect to claim 1 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
In response to the art rejection(s) of the remainder of dependent claims are rejected under 35 U.S.C 103, in case said claims are correspondingly discussed and/or argued for at least the same rationale presented in Remarks filed 06/15/2026, Examiner respectfully notes as follows. For completeness, should the mentioned claims be likewise traversed for similar reasons to independent claims 1 and 14 correspondingly, Examiner respectfully directs Applicant to the same previous supra reasons provided in the response directed towards claims 1 and 14 correspondingly discussed above. For at least the same supra provided reasons, Examiner likewise respectfully disagrees, and Applicant's arguments have been fully considered but they are not persuasive.
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1-3 and 7-15 are rejected under 35 U.S.C. 103 as being unpatentable over Borgstrom et. al. , US PgPub 2021/0074282 in view of Ali et. al. , US Patent 10,224,053 (cited in IDS).
Regarding claim 1, Borgstrom teaches computer-implemented method of audio processing, the method comprising: generating , by a machine learning model implemented via a central processing unit (CPU), first band gains and a voice activity detection value of an audio signal (see Borgstrom, Fig. 1B, the system 100 using a DNN 120 speech detection classifier (machine learning model); Borgstrom [0069] FIG. 8C shows the speech detection probabilities 830 output from the DNN(machine learning model) 120 (vad) and ( first Gain)); generating a background noise estimate based on the first band gains and the voice activity detection value (see Borgstrom, [0069] The speech detection probabilities 830 are input to the noise estimator 120 along with the spectrogram 410 by a processor system); generating second band gains by processing the audio signal (see Borgstrom, [0069] FIG. 8D shows the resultant estimated noise 840. The resultant estimated noise 840 includes both stationary and non-stationary noise components. ); generating, by the CPU, combined gains by combining the first band gains, generated by the machine learning model and the second band gains(see Borgstrom, [0069] The gain estimator 150 then generates a gain mask 850, as shown in FIG. 8E , FIG. 8C shows the speech detection probabilities 830 output from the DNN(machine learning model) 120 (vad) and ( first Gain));); and generating, by the CPU, a modified audio signal by modifying the audio signal using the combined gains ( see Borgstrom, [0069] for use by the output processor 160 for generating the enhanced spectrogram 899 of FIG. 8F as an output).
However, Borgstrom fails to teach generating, by a Wiener filter implemented via the CPU, a background noise estimate based on band energies of frames identified as noise frames using both the first band gains and the voice activity detection value.
However, Ali teaches generating , first band gains and a voice activity detection value of an audio signal (see Ali, col 8 lines 29-38 describes estimating the first gains of the signal as indicated in Ali Fig 2. Steps 210, 215, 220 ); generating, by a Wiener filter implemented via the CPU, a background noise estimate based on band energies of frames identified as noise frames using both the first band gains and the voice activity detection value( see Ali, col 12 lines 1-7 teaches determining noise detection based on VAD in step 230 which is used to determine the noise estimate); generating, by the Wiener filter, second band gains by processing the audio signal controlled by the background noise estimate(see Ali, col 11 line 62 -68 Wiener filter to determine reference noise estimation(second band gains)); generating, by the CPU, combined gains by combining the first band gains generated by the machine learning model and the second band gains generated by the Wiener filter(see Ali, Fig. 2, the gains from steps 230, 240 are processed to compute the posteriori SNR (step 245) and a priori SNR (step 250) as indicated in equations 8-9); and generating, by the CPU, a modified audio signal by modifying the audio signal using the combined gains(see Ali, col 12, lines 7-34 describes the process of conversion of filtered signal to time domain).
Borgstrom teaches the using a machine learning model to generate the filter gain function to generate enhanced speech using neural networks. Ali teaches procedure for audio signal quality enhancement based on quantitative SNR analysis and adaptive Wiener filtering. Using the known technique of wiener filtering as taught by Ali, to provide the combing of band gains in the reference Borgstrom and to provide enhanced audio signal, such as improved technical implementation using machine learning would have been obvious to one of ordinary skill in the art.
Regarding claim 2, Borgstrom in view Ali teaches the method of claim 1. Borgstrom further teaches wherein the machine learning model is generated using data augmentation to increase variety of training data (see Borgstrom, [0019-0020] training data created or augmented by human speech conversations containing noise were created by mixing the speech data with a noise signal created with at least one of a background noise data, a music data, or a non-stationary noise data, silence, reverberant speech).
Regarding claim 3, Borgstrom in view Ali teaches the method of claim 1. Borgstrom further teaches wherein generating the first band gains includes limiting the first band gains using at least two different limits for at least two different bands (see Borgstrom, [0045] Accordingly, the output from the DNN 120 can be a mask of posterior probabilities (e.g., the statistical probability that a given frequency band of a given frame contains speech) of active speech presence. The noise estimator 130 can estimate nose recursively such that the noise estimator 130 can generate a noise estimate using the output from the DNN 120, track the initial spectrum 110 when speech is not present, and smooth and/or attenuate a previous noise estimate when speech is present(different gains for different bands based on presence of speech)).
Regarding claim 7, Borgstrom in view Ali teaches the method of claim 1. Borgstrom further teaches multiplying the first band gains and the second band gains (see Borgstrom, [0045] the SNR estimator 140 can then calculate the a posteriori and a priori SNRs for the detected noise for use by the gain estimator 150, which can generate a multiplicative gain mask for use in modifying the initial spectrum 110 to suppress the detected noise without attenuating the detected speech. By applying the mask of posterior probabilities of active speech presence to the initial spectrum 110 before calculating noise estimates, the resulting gain mask more accurately suppresses noise and results in less residual noise in the enhanced spectrum 199); limiting the combined (see Borgstrom, [0052] discusses the tracking the noise and estimating a noise estimating with a smoothing factor to avoid noise filtering artifacts).
Regarding claim 8, Borgstrom in view Ali teaches the method of claim 1. Borgstrom further teaches wherein generating the modified audio signal includes modifying an amplitude spectrum of the audio signal using the combined (see Borgstrom, [0045], The SNR estimator 140 can then calculate the a posteriori and a priori SNRs for the detected noise for use by the gain estimator 150, which can generate a multiplicative gain mask for use in modifying the initial spectrum 110 to suppress the detected noise without attenuating(modifying the amplitude spectrum) the detected speech. By applying the mask of posterior probabilities of active speech presence to the initial spectrum 110 before calculating noise estimates, the resulting gain mask more accurately suppresses noise and results in less residual noise in the enhanced spectrum 199).
Regarding claim 9, Borgstrom in view Ali teaches the method of claim 1. Ali further teaches receiving an input audio signal (see Ali, Fig. 2 205 input signal); applying an overlapped window to an input audio signal to generate a plurality of frames, wherein the audio signal corresponds to the plurality of frames ( see Ali, col 11 lines 7-12 describes processing of audio signal by segmenting and computing FFT ). The motivation to combine as claim 1 applies here.
Regarding claim 10, Borgstrom in view Ali teaches the method of claim 1. Borgstrom further teaches performing spectral analysis on the audio signal to generate a plurality of bin features and a fundamental frequency of the audio signal, wherein the first band gains and the voice activity detection value are based on the plurality of bin features and the fundamental frequency (see Borgstrom, [0009-0011, 0013], discusses processing per frequency band specific and frame specific information, ; Borgstrom, [0045, 0050] discusses from a time-domain signal 101, a series of frames of pseudo-stationary segments and then apply the Fourier transform of each segment to generate the local spectra for each frame(to determine bin features and fundamental frequency features). Together, the frames can comprise the initial spectrum 110 to be provided to the DNN 120 and DNN classification for each bin (individual frequency band)).
Regarding claim 12, Borgstrom in view Ali teaches the method of claim 1. Borgstrom further teaches wherein the combined gains are combined band gains that are associated with a plurality of bands of the audio signal(see Borgstrom, [0017] discusses computing gain masks for different bands of audio signal ), the method further comprising: converting the combined band gains to combined bin gains, wherein the combined bin gains are associated with a plurality of bins (see Borgstrom, [0017) the noise estimator can be configured to calculate SNRs on a per-frame and per-frequency band basis, and the gain estimator can be configured to receive the initial spectrum, the noise variance estimate, and the SNRs.).
Regarding claim 13, is directed to a non-transitory computer readable medium claim corresponding to the method claim presented in claim 1 and is rejected under the same grounds stated above regarding claim 1.
Regarding claim 14, is directed to an apparatus claim corresponding to the method claim presented in claim 1 and is rejected under the same grounds stated above regarding claim 1.
Regarding claim 15, Borgstrom in view Ali teaches the apparatus of claim 14. Borgstrom further teaches wherein at least one limit is applied when generating at least one of the first band gains and the second band gains (see Borgstrom, [0013]the system can include a filter configured to apply a passband (limit) to the input signal. The passband can have a frequency range that corresponds to human speech. In at least some such instances, the filter can be configured to apply cepstral mean subtraction on at least one cepstral coefficient of the input signal)
Claim 4-5 is rejected under 35 U.S.C. 103 as being unpatentable over are rejected under 35 U.S.C. 103 as being unpatentable over Borgstrom et. al. , US PgPub 2021/0074282 in view of Ali et. al. , US Patent 10,224,053 (cited in IDS) further in view of A. Biswas, et. al., "Acoustic feature extraction using ERB like wavelet sub-band perceptual Wiener filtering for noisy speech recognition," 2014 Annual IEEE India Conference (INDICON), Pune, India, 2014, pp. 1-6.
Regarding claim 4, Borgstrom in view of Ali teaches the method of claim 1. However, Borgstrom in view of Ali fails to teach wherein generating the background noise estimate is based on a number of noise frames exceeding a threshold for a particular band.
However, Biswas teaches wherein generating the background noise estimate is based on a number of noise frames exceeding a threshold for a particular band (see Biswas, sect II, Then for each decomposed sub-band auditory masking threshold is calculated. This threshold is used as frequency gain function of the wiener filter to enhanced the noisy speech spectrum ).
Borgstrom, Ali and Biswas are considered to be analogous to the claimed invention because they relate to enhancing speech signals. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Borgstrom in view of Ali using a machine learning model to generate the filter gain function to generate enhanced speech using neural networks with the Noise estimation teachings of Biswas to enhance the noisy speech signal (see Biswas, sect I).
Regarding claim 5, Borgstrom in view Ali teaches the method of claim 1. However, Borgstrom in view Ali fail to teach wherein generating the second band gains includes using the Wiener filter based on a stationary noise level of a particular band. However, Biswas teaches wherein generating the second band gains includes using the Wiener filter based on a stationary noise level of a particular band (see Biswas, sect II, IIC : Non-uniform WP sub-bands decomposition of the noisy speech v(t)=s(t)+n(t)) is carried out using auditory ERB scale [4], [12], where s(t) is the clean signal, n(t) is the additive noise and t=0,1,…,N−1 is the time index. Auditory masking threshold is calculated by processing each wavelet sub-bands.. Fast Fourier transform (FFT) is applied to the output of the each WP decomposed subband output yi,k(t) to obtain noisy speech spectrum Yi,k(v). Then Wiener filter Wi,k(v) (of the particular band) is then applied to each sub-band to produce enhanced power spectrum S~i,k(v).). The same motivation to combine as claim 4 applies here.
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over are rejected under 35 U.S.C. 103 as being unpatentable over Borgstrom et. al. , US PgPub 2021/0074282 in view of Ali et. al. , US Patent 10,224,053 (cited in IDS) further in view of Every et. al. US PgPub. 2011/0257967.
Regarding claim 6, Borgstrom in view Ali teaches the method of claim 1. However, Borgstrom in view Ali fails to teach wherein generating the second band gains includes limiting the second band gains using at least two different limits for at least two different bands. However, Every teaches wherein generating the second band gains includes limiting the second band gains using at least two different limits for at least two different bands (see Every, [0089] the wiener gain values from the wiener filter module 400 are also provided to the optional mask smoother module 402. The mask smoother module 402 performs temporal smoothing of the wiener gain values, which helps to reduce the musical noise. The wiener gain values may change quickly (e.g. from one frame to the next) and speech and noise estimates can vary greatly between each frame. Thus, the use of the wiener gain values, as is, may result in artifacts (e.g. discontinuities, blips, transients, etc.). Therefore, optional filter smoothing may be performed in the mask smoother module 402 to temporally smooth the Wiener gain value)).
Borgstrom, Ali and Every are considered to be analogous to the claimed invention because they relate to enhancing speech signals. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Borgstrom in view of Ali using a machine learning model to generate the filter gain function to generate enhanced speech using neural networks with the Wiener gain optimization teachings of Every to improve user’s perception(see Every, [0007]).
Claims 11 and 16 rejected under 35 U.S.C. 103 as being unpatentable over are rejected under 35 U.S.C. 103 as being unpatentable over Borgstrom et. al. , US PgPub 2021/0074282 in view of Ali et. al. , US Patent 10,224,053 (cited in IDS) further in view of Kim et. al. US PgPub. 2014/0337021.
Regarding claim 11, Borgstrom in view Ali teaches the method of claim 10. Borgstrom teaches generating a plurality of band features based on the plurality of bin features (see Borgstrom, [0009-0011, 0013], discusses processing per frequency band specific and frame specific information), wherein the first band gains and the voice activity detection value are based on the plurality of band features and the fundamental frequency(see Borgstrom, [0045, 0050] discusses from a time-domain signal 101, a series of frames of pseudo-stationary segments and then apply the Fourier transform of each segment to generate the local spectra for each frame(to determine bin features and fundamental frequency features). Together, the frames can comprise the initial spectrum 110 to be provided to the DNN 120 and DNN classification for each bin (individual frequency band) ). However, Borgstrom in view Ali fail to teach wherein the plurality of band features are generated using one of Mel-frequency cepstral coefficients and Bark-frequency cepstral coefficients.
However, Kim teaches wherein the plurality of band features are generated using one of Mel-frequency cepstral coefficients and Bark-frequency cepstral coefficients(see Kim, [0082] discusses spectral features like Mel frequency spectrum information; [0088] discusses The subband mapper 580 may map the spectrogram 578 (e.g., power spectrogram) to a group of subbands 582 with center frequencies that are logarithmically scaled (e.g., a Bark scale). The subbands 582 may be provided to the stationarity detector 584.).
Borgstrom, Ali and Kim are considered to be analogous to the claimed invention because they relate to enhancing audio signals. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Borgstrom in view of Ali using a machine learning model to generate the filter gain function to generate enhanced speech using neural networks with the Noise characteristic and suppression teachings of Kim to enhance the noisy speech signal (see Kim, sect [0006]).
Regarding claim 16, Borgstrom in view Ali teaches the apparatus of claim 14. However, Borgstrom in view Ali fail to teach generate, by the Wiener filter, the second band gain for each of a plurality of bands, wherein the second band gain for a band is a computed Wiener gain when a number of noise frames identified for that band exceeds a threshold, and wherein the second band gain for a band is a unity gain when the number of noise frames identified for that band does not exceed the threshold.
However, Kim teaches teach generate, by the Wiener filter, the second band gain for each of a plurality of bands, wherein the second band gain for a band is a computed Wiener gain when a number of noise frames identified for that band exceeds a threshold (see Kim, [0032] teaches if the number of frames detected as including stationary noise within the stationary noise detection time interval is greater than a stationary noise detection threshold, the stationary noise detector 110 may indicate that the noise in the input audio 104 is stationary noise.[0041] The noise reference 118 may be a signal or indicator that indicates the noise to be suppressed in the input audio 104. For example, the noise reference 118 may be utilized by the noise suppressor 120 (e.g., a Wiener filter) to suppress noise in the input audio 104), and wherein the second band gain for a band is a unity gain when the number of noise frames identified for that band does not exceed the threshold(see Kim, [0066] discusses spatial noise reference to indicate to the noise suppressor ( unity gain)). The motivation to combine as noted for claim 11 applies here.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Lou, US PgPub. 2016/0240210 teaches a methods for enhancing speech signal intelligibility and for bettering performance of automatic speech recognition processes, for a speech signal in a noisy environment (see Lou, Abstract).
Z. Cui and C. Bao, "Linear Prediction-based Part-defined Auto-encoder Used for Speech Enhancement," ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Brighton, UK, 2019, pp. 6880-6884, teaches the linear prediction (LP) model-based PAE is used to predict the AR-Wiener filter parameters, in which the line spectral frequency (LSF) parameters, AR gains are used for middle feature spaces target, and Wiener mask are used for final target. Furthermore, a self-defined the LP layer that can transfer the LP parameters into magnitude spectrum is used to drive the AR-Wiener filter for obtaining magnitude spectrum of clean speech(see Cui, section I) .
Abd El-Fattah, Marwa A., et al. "Speech enhancement with an adaptive Wiener filter." International Journal of Speech Technology 17.1 (2014): 53-64 teaches an adaptive Wiener filtering method for speech enhancement (see Abd, sect. 1).
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NANDINI SUBRAMANI whose telephone number is (571)272-3916. The examiner can normally be reached Monday - Friday 12:00pm - 5:00 pm EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Bhavesh M Mehta can be reached at (571)272-7453. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/NANDINI SUBRAMANI/ Examiner, Art Unit 2656
/BHAVESH M MEHTA/ Supervisory Patent Examiner, Art Unit 2656
Prosecution Insights