Prosecution Insights
Last updated: November 20, 2026
Application No. 19/121,280

METHOD FOR ADVERSARIAL TRAINING FOR UNIVERSAL SOUND SEPARATION

Non-Final OA §102
Filed
Apr 15, 2025
Priority
Oct 17, 2022 — ES P202230890 +3 more
Examiner
SAINT CYR, LEONARD
Art Unit
2658
Tech Center
2600 — Communications
Assignee
Dolby International AB
OA Round
1 (Non-Final)
78%
Grant Probability
Favorable
1-2
OA Rounds
1y 5m
Est. Remaining
95%
With Interview

Examiner Intelligence

Grants 78% — above average
78%
Career Allowance Rate
916 granted / 1182 resolved
+15.5% vs TC avg
Strong +18% interview lift
Without
With
+17.9%
Interview Lift
resolved cases with interview
Typical timeline
3y 1m
Avg Prosecution
24 currently pending
Career history
1221
Total Applications
across all art units

Statute-Specific Performance

§101
16.9%
-23.1% vs TC avg
§103
41.7%
+1.7% vs TC avg
§102
27.2%
-12.8% vs TC avg
§112
1.3%
-38.7% vs TC avg
Black line = Tech Center average estimate • Based on career data from 1182 resolved cases

Office Action

§102
DETAILED ACTION 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 . Claim Rejections - 35 USC § 102 The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action: A person shall be entitled to a patent unless – (a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention. Claims 1- 4, 8, 9, 11, 12, 15, 16, 19, 21, 24, 27, 28 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Chen et al. (Permutation Invariant Training of Generative Adversarial Network for Monaural Speech Separation, 2018). As per claim 1, Chen et al. teach a method for adversarial training of a separator for universal sound separation of an audio mixture m of arbitrary sound sources Sk=1,...,K, the method comprising: training a context-based discriminator configured to provide a context-based loss cue based on a consideration of an input set of separated sound sources; and training the separator to minimize a loss based on the context-based loss cue provided by the context-based discriminator (“The generator network learns to model labeled data, e.g. the mapping from noisy speech samples to their clean counterparts, while the discriminator, usually a binary classifier, learns to discriminate between generated samples and target samples from training data. This framework is analogous to a two-player adversarial game, where minimax is a proven strategy. The key idea of GANs is to use the discriminator to shape the loss function of the generator”; pages 302, paragraph 2); wherein training the context-based discriminator comprises maximizing a loss based on a set of ground-truth sound sources and a fake set of separated sound sources, wherein the fake set of separated sound sources is sorted to match an order of the set of ground-truth sound sources, and wherein the fake set of separated sound sources comprises sources corresponding to separated sound sources estimated by the separator and further comprises one or more ground-truth sound sources of the set of ground-truth sound sources (“adversarial training, where we have another component, called the discriminator (D). D is in charge of transmitting information to Gon what is real (x ∼ pdata) and what is fake (G(z) ∼ pG), such that G can slightly correct its output waveform towards the realistic distribution, getting rid of the noisy signals as those are signaled to be fake…G is trained to minimize this objective, while D is trained to maximize it”; page 303, col.2). As per claim 2, Chen et al. further disclose the fake set of separated sound sources is obtained by: obtaining a set of separated sound sources corresponding to a set of estimated separated sound sources [s₁, sₖ] estimated by the separator from the audio mixture m; permuting the set of separated sound sources such that an order of the permuted set of separated sound sources matches an order of the set of ground-truth sound sources; and replacing one or more of the separated sound sources of the permuted set of separated sound sources with one or more ground-truth sound sources of the set of ground-truth sound sources to obtain the fake set of separated sound sources (“where φ∗ is the permutation that minimizes the utterance-level Separation error defined as…and P is the set of all S! permutations. With uPIT, the permutation corresponding to the minimum utterance-level separation Error is used for all frames in the utterance”; pages 302, 304). As per claim 3, Chen et al. further disclose the set of separated sound sources is permuted using a permutation matrix P*, wherein P* is the permutation matrix among a set of all permutation matrices P minimizing a loss between the set of ground truth sources and the set of separated sound sources permuted using P(“where φ∗ is the permutation that minimizes the utterance-level Separation error defined as…and P is the set of all S! permutations. With uPIT, the permutation corresponding to the minimum utterance-level separation Error is used for all frames in the utterance”; pages 302, 304). As per claim 4, Chen et al. further disclose the context-based discriminator is configured to operate in a waveform domain, a magnitude STFT domain, a filter-bank domain or a mask domain (“The STFT of waveform signal is computed with a 32-ms window and shifted every 16ms.”; page 304). As per claim 8, Chen et al. further disclose the context-based discriminator (34) is configured to operate in a waveform domain, and wherein the set of ground-truth sound sources [s₁, sₖ] and the fake set of separated sound sources [s₁, ⁵ₖ] are represented in the waveform domain (“adversarial training, where we have another component, called the discriminator (D). D is in charge of transmitting information to Gon what is real (x ∼ pdata) and what is fake (G(z) ∼ pG), such that G can slightly correct its output waveform towards the realistic distribution, getting rid of the noisy signals as those are signaled to be fake…G is trained to minimize this objective, while D is trained to maximize it”; page 303, col.2). As per claim 9, Chen et al. further disclose the fake set of separated sound sources [s₁, ⁵ₖ] is obtained by: obtaining a set of separated sound sources [s₁, sₖ] represented in the waveform domain and estimated by the separator from the audio mixture m; permuting the set of separated sound sources [s₁, sₖ] such that an order of the permuted set of separated sound sources [s*, matches an order of the set of ground- truth sound sources sₖ]; and replacing one or more of the separated sound sources Sₖ of the permuted set of separated sound sources [s₁, SK] with one or more ground-truth sound sources Sₖ of the set of ground-truth sound sources [s₁, sₖ] to obtain the fake set of separated sound sources [s₁,,,,,ₛₖ](“where φ∗ is the permutation that minimizes the utterance-level Separation error defined as…and P is the set of all S! permutations. With uPIT, the permutation corresponding to the minimum utterance-level separation Error is used for all frames in the utterance”; pages 302, 304; figs. 1, 2, and 3). As per claim 11, Chen et al. further disclose the context-based discriminator is denoted Dwave and is trained to maximize Lₑₜₓ,S = Lreal,Stx,Stx,Stx,S + Lake where is based on Dwave (m, S₁, Sₖ) and Lfake is based on Dwave (m, S₁, Sₖ)[“discriminator (D). D is in charge of transmitting information to G on what is real (x ∼ pdata) and what is fake (G(z) ∼ pG), such that G can slightly correct its output waveform towards the realistic distribution, getting rid of the noisy signals as those are signaled to be fake. In this sense, D can be understood as learning some sort of loss, for G’s output to look real.”; page 303, col.2]. As per claim 12, Chen et al. further disclose the context-based discriminator is configured to operate in a magnitude STFT domain, and wherein the set of ground-truth sound sources [|S₁|, , |Sₖ|] and the fake set of separated sound sources [|S₁|, |⁵ₖ|] are represented in the magnitude STFT domain (pages 303, col.2; 304, col.2). As per claim 15, Chen et al. further disclose the context-based discriminator is denoted DSTX,STX, StX, Stx,Stx,Stx, and is trained to maximize Lctx,S = + Lake where Lreal is based on DSTX,S |Sₖ|) and Lfake is based on DSTX,S (|M|, |S₁|, |Sₖ|), where M = STFT(m) [pages 303, col.2; 304, col.2]. As per claim 16, Chen et al. further disclose the context-based discriminator is configured to operate in a ratio mask domain, and wherein the set of ground-truth sound sources [R₁, Rₖ] and the fake set of separated sound sources [R₁, πₖ] are represented in the ratio mask domain (the model infers idea ratio mask (IRM)… D is in charge of transmitting information to G on what is real (x ∼ pdata) and what is fake (G(z) ∼ pG), such that G can slightly correct its output waveform towards the realistic distribution, getting rid of the noisy signals as those are signaled to be fake.; page 303). As per claim 19, Chen et al. further disclose training an instance-based discriminator configured to provide an instance-based loss cue based on a consideration of an individual separated sound source, wherein the loss minimized to train the separator is further based on the instance- based loss cue provided by the instance-based discriminator (“while the discriminator, usually a binary classifier, learns to discriminate between generated samples and target samples from training data. This framework is analogous to a two-player adversarial game, where minimax is a proven strategy. The key ideaof GANs is to use the discriminator to shape the loss function of the generator”; pages 302, paragraph 2). As per claim 21, Chen et al. further disclose the instance-based discriminator is configured to operate in a waveform domain, a magnitude STFT, a filter-bank domain or a mask domain (“The STFT of waveform signal is computed with a 32-ms window and shifted every 16ms.”; page 304). As per claim 24, Chen et al. further disclose the instance-based discriminator and the context-based discriminator are jointly trained (pages 302, paragraph 2). As per claim 27, Chen et al. further disclose a neural network-based system for universal sound separation of an audio mixture m of arbitrary sound sources Sk=1,...,K , wherein the system is configured to perform the method according to claim 1 (“To better explore time dependencies, recurrent neural networks (RNNs) with long short-term memory (LSTM) are adopted for both generator and discriminator”; page 302). As per claim 28, Chen et al. further disclose a method for universal sound separation, comprising: estimating by a separator a set of separated sound sources [ˢ₁,...,ₛₖ] from an audio mixture m of arbitrary sound sources Sk=1,...,K, wherein the separator is trained using the method according to claim 1 (pages 302, 303). Allowable Subject Matter Claims 5, 6, 13, 17, and 22 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims. The following is a statement of reasons for the indication of allowable subject matter: As to claims 5, 6, the prior art made of record does not teach or suggest the context-based discriminator is a first context-based discriminator configured to operate in a first domain and the method further comprises training a second context-based discriminator configured to operate in a second domain different from the first domain and to provide a second context-based loss cue based on a consideration of an input set of separated sound sources represented in the second domain, wherein the loss minimized to train the separator is further based on the context- based loss cue provided by the second context-based discriminator. As to claim 13, the prior art made of record does not teach or suggest the fake set of separated sound sources [|S₁|, |Sₖ|] is obtained by: obtaining a set of separated sound sources [|ˢ₁|, |$ₖ|] represented in the magnitude STFT domain and corresponding to a set of estimated separated sound sources [s₁, , sₖ] estimated by the separator (30) from the audio mixture m; permuting the set of separated sound sources [|$₁| |ˢₖ|] such that an order of the permuted set of separated sound sources [|ˢ₁|, |S*|] matches an order of the set of ground- truth sound sources [|S₁|, |Sₖ|]; replacing one or more of the separated sound sources |Sₖ| of the permuted set of separated sound sources [|ˢ₁|, |Sₖ|] with one or more ground-truth sound sources |Sₖ of the set of ground-truth sound sources [|S₁|, |Sₖ|] to obtain the fake set of separated sound sources [|S₁|, wherein Sₖ and Sₖ | denote the magnitude STFTs of the ground-truth sound sources Sₖ and the estimated separated sound sources Sₖ, respectively. As to claim 17, the prior art made of record does not teach or suggest the fake set of separated sound sources [|R₁|, |Rₖ|] is obtained by: obtaining a set of separated sound sources [R₁, Řₖ] represented in the ratio mask domain and corresponding to a set of estimated separated sound sources [s₁, ₛₖ] estimated by the separator (30) from the audio mixture m; permuting the set of separated sound sources [R₁,...,Řₖ] such that an order of the permuted set of separated sound sources [R₁, R*] matches an order of the set of ground- truth sound sources [R₁, , Rₖ]; replacing one or more of the separated sound sources Rₖ of the permuted set of separated sound sources [R₁, Rₖ] with one or more ground-truth sound sources Rₖ of the set of ground-truth sound sources [R₁, Rₖ] to obtain the fake set of separated sound sources [R₁, Rₖ]. As to claim 22, the prior art made of record does not teach or suggest the instance-based discriminator is a first instance-based discriminator configured to operate in a first domain and the method further comprises training a second instance-based discriminator configured to operate in a second domain different from the first domain and to provide a second instance-based loss cue based on a consideration of an individual separated sound source represented in the second domain, wherein the loss minimized to train the separator is further based on the second instance-based loss cue provided by the second instance-based discriminator. Conclusion The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Wang et al. teach Targeted Voice Separation by Speaker For Speech Recognition. Xiao et al. teach DENOISING DIFFUSION GENERATIVE ADVERSARIAL NETWORKS. Zeghidour et al. teach Separating Speech By Source In Audio Recordings By Predicting Isolated Audio Signals Conditioned On Speaker Representations. Mesgarani et al. teach Methods For Speech Separation And Neural Decoding Of Attentional Selection In Multi-speaker Environments. Any inquiry concerning this communication or earlier communications from the examiner should be directed to LEONARD SAINT-CYR whose telephone number is (571)272-4247. The examiner can normally be reached Monday- Friday. 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, Richemond Dorvil can be reached at (571)272-7602. 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. /LEONARD SAINT-CYR/Primary Examiner, Art Unit 2658
Read full office action

Prosecution Timeline

Apr 15, 2025
Application Filed
Sep 10, 2026
Non-Final Rejection mailed — §102
Nov 11, 2026
Interview Requested

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12801867
SWITCHBOARD PLATFORM FOR FOUNDATION MODELS
2y 6m to grant Granted Nov 17, 2026
Patent 12784748
Delay Gating for Voice Arbitration
2y 9m to grant Granted Nov 03, 2026
Patent 12784120
ADAPTIVE MISSION STRATEGY OPTIMIZATION THROUGH AI-GENERATED AMBIENT RECOMMENDATIONS FOR TACTICAL COMMAND
2y 1m to grant Granted Nov 03, 2026
Patent 12755734
INSERTION ERROR REDUCTION WITH CONFIDENCE SCORE-BASED WORD FILTERING
3y 6m to grant Granted Oct 06, 2026
Patent 12755095
PERCEPTION CAPTURE SYSTEM
2y 4m to grant Granted Oct 06, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

1-2
Expected OA Rounds
78%
Grant Probability
95%
With Interview (+17.9%)
3y 1m (~1y 5m remaining)
Median Time to Grant
Low
PTA Risk
Based on 1182 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month