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.
Claim(s) 1-7,9,12-19,21 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Berol (US 11869531 B1).
As per claim 1, an audio recognition method, comprising:
obtaining a target feature map of audio data based on a multi-level feature map of the audio data (para. 60: A stripe window slides over the high-level feature, where the processing based on the window forms the target feature map));
determining a feature representation of the audio data based on the target feature map (any of the processing cited in para 60 or 61, such as the audio event predictions and or regions of interest on the high level feature map, and or the probability determinations and or regression or nms processes per para 61); and
determining a recognition result for the audio data at least based on the feature representation (per para 65, any of the processes supporting: for the final classifier to predict audio events from event proposals. R-CRNN can be extended to detect different classes of events simultaneously within the same framework by adding more outputs to the final classifier).
As per claim 12. An electronic device, comprising: a processor; and a memory coupled to the processor, wherein the memory has stored therein instructions (the system of the claim 1 rejections requires processor memory and software to be implemented) that, when executed by the processor, cause the electronic device to perform actions comprising:
Obtaining a target feature map of audio data based on a multi-level feature map of the audio data; determine a feature representation of the audio data based on the target feature map; and
determining a recognition result for the audio data at least based on the feature representation.
(per the claim 1 rejection).
As per claim 13, (Currently amended) A computer program product tangibly stored on a computer-readable medium and comprising machine-executable instructions (per the claim 12 rejection) that, when executed, cause a machine to; obtain a target feature map of audio data based on a multi-level feature map of the audio data;determine a feature representation of the audio data based on the target feature map; and determine a recognition result for the audio data at least based on the feature representation. (per the claim 1 and 12 rejections)
As per claims 2,14, the method according to claim 1, wherein obtaining the target feature map comprises:
obtaining the multi-level feature map of the audio data (running the window over the high level feature map on a frame by frame basis),
wherein a next-level feature map in the multi-level feature map is extracted from a previous-level feature map (the results of the window processing are sliding per para 60 which means the window contains current and previous level feature map per para 60 ); and
performing feature reconstruction at least based on the next-level feature map and the previous-level feature map, to determine the target feature map (the processing of para 60-65 via the use of a sliding window and also via a recurrent learning layer pern para 60, which by definition facilitates feature reconstruction based on current and previous feature level maps).
As per claims 3,15, the method according to claim 2, wherein the multi-level feature map comprises at least:
a first-level feature map extracted from the audio data (one of the regions per para 61: At each frame of the high-level feature map, multiple regions of different sizes center around it); and
a second-level feature map extracted based on the first-level feature map (another one of said regions).
As per claims 4,16, the method according to claim 3, wherein the feature reconstruction comprises at least:
expanding the second-level feature map into a first-level spare feature map (the lower dimensional result via the window per para 60); and
determining the target feature map based on the first-level spare feature map and the first- level feature map (both of the elements are used in producing the output of the regionalized feature map per para 60-65).
As per claims 5,17. the method according to claim 1,
wherein the audio data is training data (a neural network processing per para 60-65 trains/learns over time with its data/audio data), and the method further comprises:
determining a loss function value of a trained recognition model based on the recognition result (the loss function per para 63) and a pre-labeled ground-truth result for the training data (para 63), to update parameters of the recognition model (para 63-65 noting the neural network and sliding window and frame based processing ).
As per claims 6,18, the method according to claim 5, further comprising:
determining a distribution of feature representations corresponding to audio clips that fall into or do not fall into a classification of chorus (the output feature map per the processing of para 60-65 to make an audio prediction noting the context of a song per para 152); and
determining sampled feature representations in the distribution as additional feature representations (para. 61: and another dense layer (reg) to encode the coordinates of interval proposals (2 k coordinates). Following the settings outlined above, these k proposals are parameterized by shifting and scaling relative to k anchor intervals).
As per claims 7,19, the method according to claim 6, wherein determining the sampled feature representations as the additional feature representations comprises:
sampling a predetermined number of feature representations in the distribution (the samples in the window and or in each frame), and
using the predetermined number of feature representations as the additional feature representations (para 60: The size of sliding window is 3*n, where n is the height of the high-level feature map (n=2U in our case). The receptive field of the sliding window is 557 ms (3×8×frame shift), ).
As per claims 9,21, the method according to claim 6, wherein determining the recognition result at least based on the feature representation comprises: inputting the feature representation and the additional feature representations into a fully connected layer of the recognition model, to determine the recognition result (the neural network based processing cited above comprises fully connected layers for each stage noting para 49, including a DNN: deep neural network (“DNN”). Information from one layer can be processed and provided to a next layer) (where each of the elements as per para 49-51, including the feature representations and additional feature representations must be implemented into a fully connected layer in order to function with the neural network as disclosed).
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 8,20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Berol (US 11869531 B1) as applied to claim 1, 7,18,19 above, and further in view of Azernikov et al (US 20220215531 A1).
As per claims 8,20, Berol discloses the method according to claim 7, but does not specify wherein determining the loss function value comprises: determining an upper limit of a loss function of the recognition model by setting the predetermined number to positive infinity, to determine the loss function value.
Azernikov teaches that neural network based classifying can include determining an upper limit of a loss function of the recognition model by setting the predetermined number to positive infinity (para 172). Azernikov teaches per para 169 that: the first segment path 2053 and second segment path 2056 can be determined using Dijkstra's shortest path algorithm. It would have been obvious to one skilled in the art to implement a well known algorithm comprising the infinity based processing cited above in the recognition process of Berol for the purpose of conforming to well known protocols and standards.
Claim(s) 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over Berol (US 11869531 B1) as applied to claim 1, 7 above, and further in view of Morsy et al (US 20220215531 A1).
As per claim 10, Berol discloses the method according to claim 1 including processing songs, but does not specify wherein the audio data is an audio clip of a song, and determining the recognition result for the audio data comprises: determining that the audio clip falls into a classification of chorus; or determining that the audio clip does not fall into the classification of chorus.
Morsy teaches to use neural network based feature maps/decomposed signals, per para 74 to identify the parts of a song, including a chorus (para 81,83). Morsy teaches that this helps to provide suitable mixing conditions per para 6. It would have been obvious to one skilled in the art that the neural network based classification of Berol could be used to classify choruses of songs for the purpose of providing suitable mixing ocnditions.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALEXANDER KRZYSTAN whose telephone number is 571-272-7498, and whose email address is alexander.krzystan@uspto.gov
The examiner can usually be reached on m-f 7:30-4:00 est.
If attempts to reach the examiner by telephone or email are unsuccessful, the examiner’s supervisor, Fan Tsang can be reached on (571) 272-7547.
The fax phone numbers for the organization where this application or proceeding is assigned are 571-273-8300 for regular communications and 571-273-8300 for After Final communications.
/ALEXANDER KRZYSTAN/Primary Examiner, Art Unit 2653
Examiner Alexander Krzystan
June 17, 2026
Prosecution Insights