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 .
Response to Arguments
Applicant's arguments filed 03/09/2026 have been fully considered but they are not persuasive.
In response to applicant's arguments against the references individually, one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). Particularly, Applicant argues, on pages 8 and 9 of Applicant’s Response, that Examiner incorrectly characterizes Bhamidipati as showing “transforming a word from a first portion of a first word (i.e., identifying a portion of a keyword)” because, according to Applicant, the words are stored word recordings not words generated in real-time. Examiner respectfully disagrees.
It is not Bhamidipati’s modification of words alone that teaches such a limitation. Indeed, it is the combination of Danieli and Bhamidipati. Applicant addresses Danieli on pages 8 and 9 of Applicant’s Response, however, Applicant merely states that Danieli does not teach this element. Neither Danieli or Bhamidipati are alone relied upon to teach “identifying a first portion of a keyword in speech during generation of speech.” Bhamidipati’s alteration of stored words by syllable in combination with Danieli’s live censorship of broadcasts (i.e., the generation of speech occurs via the broadcast) meets the limitations of the claims. Indeed, the processing of words by syllable of Bhamidipati in combination with Danieli’s live replacement of words in a broadcast would have been obvious to one of skill in the art because such a modification would have been a predictable variation of known solutions, due to Bhamidipati and Danieli sharing similar fields of endeavor and goals of altering words or voice signals. Furthermore, the format of the words operated on in Bhamidipati is irrelevant to the combination with Danieli because Danieli provides the teachings of processing in real-time which are combined with Bhamidipati’s teachings of editing words by syllable. As such, the 35 U.S.C. § 103 rejections laid out below are maintained for at least this reason.
Applicant further argues, on page 9 of Applicant’s Response, that Bhamidipati and Danieli do not teach transforming the keyword into a different word by introducing the waveform into the audio signal. Particularly, Applicant argues that Bhamidipati does not teach such a limitation because, according to Applicant, Bhamidipati’s process is batch transformation between two complete stored word samples. However, this fails to take the teachings of Danieli, and the combination of the two references into account. Particularly, Danieli’s live-censorship of speech by replacing words using phonemes of the speaker (Danieli at ¶ [0059]) in view of Bhamidipati’s alteration of first and second portions of words would have made it obvious to adapt Bhamidipati’s system to Danieli’s live censorship task. Furthermore, the alteration of a word, or speech in live speech transmission, as done by Danieli in view of Bhamidipati, would have resulted in the introduction of the modified waveform into the audio signal. Indeed, the modification of any word being broadcast would be a predictable variation of Danieli and Bhamidipati’s system due to the nature of live broadcasts and the need for live censorship, in addition to their combined teachings in similar fields. As such, Bhamidipati and Danieli, when combined, indeed teach transforming the keyword into a different word by introducing a waveform into an audio signal, and the 35 U.S.C. § 103 rejections laid out below are maintained for at least this reason.
Further, Applicant argues, on pages 10 – 11 of Applicant’s Response, that Feinauer, which is relied upon to demonstrate that real-time audio processing can be, and is, performed in 100 ms, has 3 issues that prevent it from teaching the limitations of the claims. These issues are:
Feinauer’s 100 ms lag is a system-wide pipeline buffer applied to the entire audio stream, not detecting of a keyword, identification, waveform determination and transformation within 200 ms tied to that keyword’s generation.
Feinauer’s system accomplishes pitch, tone, and volume transformation, not identification, waveform determination, and transformation of claim 1 according to applicant.
Feinauer’s word-level replacement requires a “model to transcribe a spoken word from speech-to-text determine the equivalent word by referencing an uploaded dialect dictionary and modulate a voice in real-time to replace the spoken word with an entirely new word using a text-to-speech model” which applicant alleges would take longer than 200 ms, and therefore Feinauer’s 100 ms buffer only supports acoustic feature shifting.
Examiner respectfully disagrees with Applicant’s assertion of such issues with Feinauer.
Regarding the first issue, of a 100 ms system-wide pipeline buffer. Such a teaching is, again, regarded in isolation of the combination with other references. Particularly, it is Danieli-Bhamidipati in view of Feinauer that achieves such a goal as Danieli operates on real-time audio (Danieli at ¶ [0029]) and Bhamidipati’s alteration waveforms (e.g., Bhamidipati at ¶¶ [0041] – [0042]) along with Danieli’s live audio censorship are akin to Feinauer’s real-time processing and alteration of words using a 100 ms buffer to accommodate for processing time to keep the system in “real-time.” As such, in live censorship, when the first syllable of the word is detected using Bhamidipati’s substituted process in Danieli’s system, Feinauer’s 100 ms demonstrates that the processing performed, i.e., alteration of waveforms, would be achieved within 100 ms (pitch, tone, volume, etc. are all waveform alterations akin to the waveform processing needed to alter the word such as in Bhamidipati). As such, the 200 ms window, alleged by applicant, would be achieved by the combination of these references occupying similar fields of endeavor.
Regarding the second issue, Feinauer is not relied upon to teach the identification, determination, or transformation of such words. Indeed, Danieli-Bhamidipati’s teachings are relied upon for such modification limitations, and their teachings, as laid out below, in combination with Feinauer’s 100 ms processing time, would have made obvious the processing and introduction of a waveform into real-time audio, preserving the real-time audio experience for users by completing waveform modification within the 100 ms window. Furthermore, Danieli, at ¶ [0059] recognizes that the modifications performed by Feinauer are performed on the real-time audio to replace words using previously uttered phonemes by the user. Particularly, Danieli teaches replacing the words “God” and “damn” with “gosh” and “darn” these words both begin with the same syllables and convey the same meaning, though less extreme. As such, a person of ordinary skill in the art would have recognized Danieli’s example to be directly achievable by Bhamidipati’s process. Further, A person of skill in the art would have recognized that attenuation and/or replacement of individual phonemes, can be achieved for real-time audio as recognized by Feinauer by processing and editing the audio stream suing continuous signal level transformation of acoustic properties. Therefore, Danieli-Bhamidipati’s system would have been obvious to implement within 100 ms of the detected expletive as this processing time would preserve real-time audio for the users.
Regarding the third issue, while Feinauer does discuss the potential for real-time word replacement to introduce lag too great for real-time communication, Feinauer only discusses such a lag as a possibility in “some implementations” not in all implementations. As such, this discussion infers that it is also possible to achieve real-time word replacement by modulating the voice as only some implementations of such a system create lag too great for real-time communication. Thus, Feinauer’s disclosure does not preclude such modification of audio signals within a “real-time” time frame. Indeed, Feinauer’s disclosure specifically references the fact that only “some implementations” create too much lag for real-time. As such, Feinauer’s full-word process does not exceed claim 1’s 200 ms and the 35 U.S.C. § 103 rejections laid out below are maintained in light of the reasons laid out above.
Finally, Applicant argues that there is not motivation to combine the cited references. Examiner disagrees. Particularly, Examiner notes that the cited references all process or alter words in broadcasts or other forms using computer systems and therefore occupy very similar fields of endeavor. Further, Examiner notes that the motivation to combine is not simply “a better user experience” as oversimplified by Applicant. Indeed, while this is a benefit provided to the user, the true benefit is the preservation of real-time communication while performing enhanced audio processing. Indeed, Feinauer’s audio processing is directly related to Danieli’s replacement of words using previously uttered phonemes or attenuation of live audio signals because both systems process live audio and substitute words and/or alter waveforms in real-time audio/broadcasts. (Danieli at ¶ [0059] and Feinauer 8:41 – 9:2). Therefore, Feinauer’s 100 ms processing time would have been a predictable application of well-known audio processing techniques within similar fields of endeavor which allowed for the preservation of real-time communication while also providing the user a better experience by maintaining real-time information transmission. This is not solely motivated by providing a better user experience. Indeed, this is merely an additional benefit provided for by the obvious combination of well-known audio processing techniques applied to exceedingly similar fields of endeavor that achieve similar results. As such, the 35 U.S.C. § 103 rejections are maintained in light of the rejections laid out below and all the reasons laid out above.
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 – 6, and 46 - 51 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication No. 2006/0095262 A1 to Damon V. Danieli (hereinafter Danieli) in view of U.S. 2011/0093270 A1 to Narayan Lakshmi Bhamidipati (hereinafter Bhamidipati) and in further view of U.S. Patent No. 11,450,311 B2 to Christoph Johan Feinauer et al. (hereinafter Feinauer).
Regarding claim 1, Danieli teaches an apparatus to modify sound of speech in an audio signal, the apparatus comprising: (Danieli teaches a system for automatically censoring a real-time audio source (i.e., modifying speech of an audio signal.) Danieli at ¶¶ [0029] - [0030]. Further, Danieli’s teachings of automatic censoring during live broadcast are applicable to any sort of broadcast of voice. For example, voice may be broadcast in telephone calls, video calls, live chat sessions, livestreams, cross-platform communications, gaming lobbies, etc.)) memory; instructions in the apparatus; (Danieli teaches the system comprising memory instructions stored in the memory. Danieli at ¶ [0013].)
and processor circuitry to execute the instructions to: (Danieli teaches the system comprising a processor that executes the instructions stored in the memory. Danieli at ¶ [0013].
Danieli, however, does not teach identifying a first portion of a keyword in the speech during generation of the speech; determining a waveform to replace a second portion of the keyword; and transforming the keyword into a different word by introducing the waveform into the audio signal.
In a similar field of endeavor (i.e., censoring media and replacing words in media) Bhamidipati teaches identifying a first portion of a keyword in the speech during generation of the speech; (Bhamidipati teaches transforming a word from a first portion of a first word. (i.e., identifying a portion of a keyword) Bhamidipati at ¶¶ [0014] - [0021].)
determining a waveform to replace a second portion of the keyword; (Bhamidipati teaches using portions of a one word to replace portions of another word where the part of the word used to replace the portion of the other word is a waveform. (i.e., the waveform is determined to replace a portion of the keyword.) Bhamidipati at ¶¶ [0036] - [0039].)
and transforming the keyword into a different word by introducing the waveform into the audio signal. (Bhamidipati teaches transforming one word using the syllables of another to form a new word using the syllables of both determined from a waveform of audio. (i.e., transforming a keyword into a different word by introducing the waveform into the audio signal.) Bhamidipati at ¶¶ [0036] - [0039] and [0051] - [0052].)
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date to combine the teachings of Danieli with the teachings of Bhamidipati to provide identifying a first portion of a keyword in the speech during generation of the speech; determining a waveform to replace a second portion of the keyword; and transforming the keyword into a different word by introducing the waveform into the audio signal. Doing so would have preserved the characteristics of the original speaker's voice by preserving properties of the original user's speech as recognized by Bhamidipati at ¶¶ [0018] – [0019].
Danieli in view of Bhamidipati (hereinafter Danieli-Bhamidipati), however, do not expressly teach wherein the identifying, determining and transforming are performed within less than 200 ms from generation of the first portion of the keyword.
In a similar field of endeavor (e.g., real-time processing and modification of voice signals), Feinauer teaches wherein the identifying, determining and transforming are performed within less than 200 ms from generation of the first portion of the keyword. (Feinauer teaches introducing a lag delay of 100 ms to allow for the processing of the words while still allowing the audio to sound real-time to the user. Feinauer at 22:60 - 23:9. Further, Feinauer teaches identifying and replacing words as part of an accent and dialect modification process (i.e., identifying the word to be replaced, determining which word/waveform to replace it, and transforming the word by replacing it.). Feinauer at 16:66 - 17:14. Therefore, a person of ordinary skill in the art would have recognized that replacement of words or portions of words can be performed within 100 ms of the speech to prevent interruption of the flow of communication.)
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date to combine the teachings of Danieli-Bhamidipati with the teachings of Feinauer (hereinafter Danieli-Bhamidipati-Feinauer) to provide the identifying, determining and transforming are performed within less than 200 ms from generation of the first portion of the keyword. Doing so would have allowed modification of a speech signal while preserving real-time transmission/communication for the users as recognized by Feinauer at 22:60 – 23:9. Further, Danieli teaches performing real-time audio processing and censorship in live chats such as video game voice sessions. Danieli at ¶ [0007]. Therefore, Feinauer’s 100 ms processing time would have been a predictable application of well-known audio processing techniques within similar fields of endeavor which allowed for the preservation of real-time communication while also providing the user a better experience by maintaining real-time information transmission. As such, a person of ordinary skill in the art would have found it obvious to apply Feinauer’s teachings to Danieli’s video game chat sessions in order to preserve real-time voice processing for a better user experience.
Regarding claim 2, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 1 as laid out above. Further, Bhamidipati teaches the apparatus of claim 1, wherein the processor circuitry is to: identify an attribute of the speech; and adjust the waveform based on the attribute. (Bhamidipati teaches determining properties (i.e., attributes) of a first word and transforming another word using the properties such that the transformed word has properties similar to the first. (i.e., the waveform is adjusted based on the attribute.) Bhamidipati at ¶¶ [0022] - [0031].)
Regarding claim 3, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 2 as laid out above. Further, Bhamidipati teaches the apparatus of claim 2, wherein the attribute is a volume. (Bhamidipati teaches the property (i.e., attribute) being loudness (volume). Bhamidipati at ¶ [0019].)
Regarding claim 4, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 2, as laid out above. Further, Bhamidipati teaches the apparatus of claim 2, wherein the attribute is a vocal register. (Bhamidipati teaches the property (i.e., attribute) being pitch (i.e., vocal register). Bhamidipati at ¶ [0019].)
Regarding claim 5, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 2, as laid out above. Further, Bhamidipati teaches the apparatus of claim 2, wherein the attribute is a prosody. (Bhamidipati teaches the property (i.e., attribute) being tempo (i.e., prosody/rhythm). Bhamidipati at ¶ [0019].)
Regarding claim 6, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 2, as laid out above. Further, Bhamidipati teaches the apparatus of claim 2, wherein the attribute is a speaking rate. (Bhamidipati teaches the property (i.e., attribute) being speed of utterance (i.e., speaking rate.) Bhamidipati at ¶ [0019].)
Regarding claim 46, Danieli teaches a non-transitory machine-readable medium comprising instructions that, when executed, cause one or more processors to at least: (Danieli teaches a system for automatically censoring a real-time audio source (i.e., modifying speech of an audio signal.) Danieli at ¶¶ [0029] - [0030]. Danieli teaches the system comprising a processor that executes instructions stored in the memory. Danieli at ¶ [0013]. Further, Danieli’s teachings of automatic censoring during live broadcast are applicable to any sort of broadcast of voice. For example, voice may be broadcast in telephone calls, video calls, live chat sessions, livestreams, cross-platform communications, gaming lobbies, etc.))
Danieli, however, does not teach identifying a first portion of a keyword in the speech during generation of the speech; determining a waveform to replace a second portion of the keyword; and transforming the keyword into a different word by introducing the waveform into the audio signal.
In a similar field of endeavor (i.e., censoring media and replacing words in media) Bhamidipati teaches identifying a first portion of a keyword in the speech during generation of the speech; (Bhamidipati teaches transforming a word from a first portion of a first word. (i.e., identifying a portion of a keyword) Bhamidipati at ¶¶ [0014] - [0021].)
determining a waveform to replace a second portion of the keyword; (Bhamidipati teaches using portions of a one word to replace portions of another word where the part of the word used to replace the portion of the other word is a waveform. (i.e., the waveform is determined to replace a portion of the keyword.) Bhamidipati at ¶¶ [0036] - [0039].)
and transforming the keyword into a different word by introducing the waveform into the audio signal. (Bhamidipati teaches transforming one word using the syllables of another to form a new word using the syllables of both determined from a waveform of audio. (i.e., transforming a keyword into a different word by introducing the waveform into the audio signal.) Bhamidipati at ¶¶ [0036] - [0039] and [0051] - [0052].)
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date to combine the teachings of Danieli with the teachings of Bhamidipati to provide identifying a first portion of a keyword in the speech during generation of the speech; determining a waveform to replace a second portion of the keyword; and transforming the keyword into a different word by introducing the waveform into the audio signal. Doing so would have preserved the characteristics of the original speaker's voice by preserving properties of the original user's speech as recognized by Bhamidipati at ¶¶ [0018] – [0019].
Danieli in view of Bhamidipati (hereinafter Danieli-Bhamidipati), however, do not expressly teach wherein the identifying, determining and transforming are performed within less than 200 ms from generation of the first portion of the keyword.
In a similar field of endeavor (e.g., real-time processing and modification of voice signals), Feinauer teaches wherein the identifying, determining and transforming are performed within less than 200 ms from generation of the first portion of the keyword. (Feinauer teaches introducing a lag delay of 100 ms to allow for the processing of the words while still allowing the audio to sound real-time to the user. Feinauer at 22:60 - 23:9. Further, Feinauer teaches identifying and replacing words as part of an accent and dialect modification process (i.e., identifying the word to be replaced, determining which word/waveform to replace it, and transforming the word by replacing it.). Feinauer at 16:66 - 17:14. Therefore, a person of ordinary skill in the art would have recognized that replacement of words or portions of words can be performed within 100 ms of the speech to prevent interruption of the flow of communication.)
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date to combine the teachings of Danieli-Bhamidipati with the teachings of Feinauer (hereinafter Danieli-Bhamidipati-Feinauer) to provide the identifying, determining and transforming are performed within less than 200 ms from generation of the first portion of the keyword. Doing so would have allowed modification of a speech signal while preserving real-time transmission/communication for the users as recognized by Feinauer at 22:60 – 23:9. Further, Danieli teaches performing real-time audio processing and censorship in live chats such as video game voice sessions. Danieli at ¶ [0007]. Therefore, Feinauer’s 100 ms processing time would have been a predictable application of well-known audio processing techniques within similar fields of endeavor which allowed for the preservation of real-time communication while also providing the user a better experience by maintaining real-time information transmission. As such, a person of ordinary skill in the art would have found it obvious to apply Feinauer’s teachings to Danieli’s video game chat sessions in order to preserve real-time voice processing for a better user experience.
Regarding claim 47, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 46 as laid out above. Further, Bhamidipati teaches the machine-readable medium of claim 46, wherein the instructions cause the one or more processors to: identify an attribute of the speech in the audio signal; and adjust the waveform based on the attribute. (Bhamidipati teaches determining properties (i.e., attributes) of a first word and transforming another word using the properties such that the transformed word has properties similar to the first. (i.e., the waveform is adjusted based on the attribute.) Bhamidipati at ¶¶ [0022] - [0031].)
Regarding claim 48, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 47 as laid out above. Further, Bhamidipati teaches the machine-readable medium of claim 47, wherein the attribute is a volume. (Bhamidipati teaches the property (i.e., attribute) being loudness (volume). Bhamidipati at ¶ [0019].)
Regarding claim 49, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 47 as laid out above. Further, Bhamidipati teaches the machine-readable medium of claim 47, wherein the attribute is a vocal register. (Bhamidipati teaches the property (i.e., attribute) being pitch (i.e., vocal register). Bhamidipati at ¶ [0019].)
Regarding claim 50, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 47 as laid out above. Further, Bhamidipati teaches the machine-readable medium of claim 47, wherein the attribute is a vocal register. (Bhamidipati teaches the property (i.e., attribute) being pitch (i.e., vocal register). Bhamidipati at ¶ [0019].)
Regarding claim 51, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 47 as laid out above. Further, Bhamidipati teaches the machine-readable medium of claim 47, wherein the attribute is a speaking rate. (Bhamidipati teaches the property (i.e., attribute) being speed of utterance (i.e., speaking rate.) Bhamidipati at ¶ [0019].)
Claims 7 – 8, and 52 - 53 are rejected under 35 U.S.C. 103 as being unpatentable over Danieli-Bhamidipati-Feinauer as applied to claims 1 and 46 above, and further in view of U.S. Patent Application Publication No. 2022/0115033 A1 to William Carter Huffman et al. (hereinafter Huffman)
Regarding claim 7, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 1 as laid out above. Danieli-Bhamidipati-Feinauer, however, does not teach the limitations of claim 7.
In a similar field of endeavor, (i.e., moderation of audio content.) Huffman teaches the apparatus of claim 1, wherein the processor circuitry is to: identify text of the different word based on the keyword; convert the text to speech; and determine the waveform based on the converted text to speech. (Huffman teaches replacing phonemes in an original speech to produce non-offensive words using a specialized text-to-speech engine tuned to the speaker's vocal cords. (i.e., the text to speech engine determines what word to replace the offensive word with and converts the text of that word to speech to replace the speaker's words.) Huffman at ¶ [0162]. As laid out above, Huffman teaches replacing words using a text-to-speech engine, and more specifically, replacing specific phonemes of the original word. Thus, replacing specific phonemes of a word using a text-to-speech engine would determine the waveform to replace the phonemes based on the text converted to speech.)
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date to combine the teachings of Danieli-Bhamidipati-Feinauer with the teachings of Huffman to provide the limitations of claim 7. Doing so would have made the system more robust and improved the system over time using training data and periodically retraining the system as recognized by Huffman at ¶ [0125].
Regarding claim 8, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 1 as laid out above. Danieli-Bhamidipati-Feinauer, however, does not teach the limitations of claim 8.
In a similar field of endeavor, (i.e., moderation of audio content.) Huffman teaches determining a source phoneme sequence of the keyword; (Huffman teaches replacing phonemes of an original speech (i.e., determining a source phoneme sequence of the keyword). Huffman at ¶ [0162].)
Identifying a target phoneme sequence based on the source phoneme sequence; (Huffman teaches replacing offensive language using specific phonemes from the source speech using a text-to-speech engine. (i.e., identifying target phoneme or offensive language and replacing it using generated phonemes matching the speaker’s voice.) Huffman at ¶ [0162].) and
Building the waveform based on the target phoneme sequence. (Huffman teaches replacing the phonemes of a piece of offensive language using a text-to-speech engine that generates a replacement for the offensive language (e.g., target phoneme). Huffman at ¶ [0162]. As such, replacing a phoneme with generated speech is building a waveform (i.e., generated speech is a waveform) based on the target phoneme sequence (e.g., replacing offensive language with generated speech where specific phonemes of the offensive language are replaced is generating the speech based on the targeted offensive language/phonemes.))
Regarding claim 52, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 46 as laid out above. Danieli-Bhamidipati-Feinauer, however, does not teach the limitations of claim 52.
In a similar field of endeavor, (i.e., moderation of audio content.) Huffman teaches the machine-readable medium of claim 46, wherein the instructions cause the one or more processors to: identify text of the different word based on the keyword; convert the text to speech; and determine the waveform based on the converted text to speech. (Huffman teaches replacing phonemes in an original speech to produce non-offensive words using a specialized text-to-speech engine tuned to the speaker’s vocal cords. (i.e., the text to speech engine determines what word to replace the offensive word with and converts the text of that word to speech to replace the speaker’s words.) Huffman at ¶ [0162]. As laid out above, Huffman teaches replacing words using a text-to-speech engine, and more specifically, replacing specific phonemes of the original word. Thus, replacing specific phonemes of a word using a text-to-speech engine would determine the waveform to replace the phonemes based on the text converted to speech.)
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date to combine the teachings of Danieli-Bhamidipati-Feinauer with the teachings of Huffman to provide the limitations of claim 7. Doing so would have made the system more robust and improved the system over time using training data and periodically retraining the system as recognized by Huffman at ¶ [0125].
Regarding claim 53, Danieli-Bhamidipati-Feinauer teaches all the limitations of claim 46 as laid out above. Danieli-Bhamidipati-Feinauer, however, does not teach the limitations of claim 53.
In a similar field of endeavor, (i.e., moderation of audio content.) Huffman teaches determining a source phoneme sequence of the keyword; (Huffman teaches replacing phonemes of an original speech (i.e., determining a source phoneme sequence of the keyword). Huffman at ¶ [0162].)
Identifying a target phoneme sequence based on the source phoneme sequence; (Huffman teaches replacing offensive language using specific phonemes from the source speech using a text-to-speech engine. (i.e., identifying target phoneme or offensive language and replacing it using generated phonemes matching the speaker’s voice.) Huffman at ¶ [0162].) and
Building the waveform based on the target phoneme sequence. (Huffman teaches replacing the phonemes of a piece of offensive language using a text-to-speech engine that generates a replacement for the offensive language (e.g., target phoneme). Huffman at ¶ [0162]. As such, replacing a phoneme with generated speech is building a waveform (i.e., generated speech is a waveform) based on the target phoneme sequence (e.g., replacing offensive language with generated speech where specific phonemes of the offensive language are replaced is generating the speech based on the targeted offensive language/phonemes.))
Claims 9 – 10, and 54 – 55 are rejected under 35 U.S.C. 103 as being unpatentable over Danieli-Bhamidipati-Feinauer in view of Huffman (hereinafter Danieli-Bhamidipati-Feinauer-Huffman) as applied to claims 7 – 8, and 52 – 53, above, and further in view of 2020/0243094 A1 to David Thomson et al. (hereinafter Thomson).
Regarding claim 9, Danieli-Bhamidipati-Feinauer-Huffman teaches all the limitations of claim 8 as laid out above. Danieli-Bhamidipati-Feinauer-Huffman, however, does not teach the limitations of claim 9.
In a similar field of endeavor, (i.e., speech recognition and profanity filtering.) Thomson teaches the apparatus of claim 8, wherein the processor circuitry is to implement a neural network to maintain characteristics of a voice speaking the keyword in the speech signal with the different word. (Thomson teaches using neural networks trained on a speaker profile using speech patterns of the speaker. (i.e., the model is trained to maintain the patterns of the original speaker). Thomson at ¶ [0128]. As such, Thomson in view of Huffman's specialized text-to-speech model tuned to the speaker’s vocal cords would be a neural network maintaining characteristics of the original speaker when generating new speech to replace the original speech.)
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date to combine the teachings of Danieli-Bhamidipati-Feinauer-Huffman with the teachings of Thomson to provide the limitations of claim 9. Doing so would have improved the accuracy of the ASR system as recognized by Thomson at ¶ [0129]. Further, As Thomson recognizes the functionality of profanity filtering and employing the invention of Thomson in profanity filtering it would have been obvious to combine Thomson with Danieli-Bhamidipati-Feinauer-Huffman to provide the limitations of claim 9.
Regarding claim 10, Danieli-Bhamidipati-Feinauer-Huffman in view of Thomson (hereinafter Danieli-Bhamidipati-Feinauer-Huffman-Thomson) teaches all the limitations of claim 9 as laid out above. Further, Huffman teaches the apparatus of claim 9, wherein the processor circuitry is to: disentangle characteristics of the voice; (Huffman teaches using a separate independent neural network to extract information from speech that is not part mapping the voice based on characteristics such as timbre or pitch. (i.e., the characteristics of the voice, timbre and pitch, are disentangled when the neural network extracts information separately from the voice characteristics.) Huffman at ¶ [0063].)
learn representations of the speech in the audio signal independent of the source phoneme sequence; (Huffman teaches forming representations of speech using voice mappings. (i.e., learning representations of speech without using phoneme sequences, instead using voice mappings.) Huffman at ¶ [0063].)
Further, Danieli teaches building the waveform based on the learned representations. (Danieli teaches replacing a word derived from the speech previously uttered by the speaker. Danieli at ¶ [0055] - [0059]. As such, Danieli's replacement of a profane word spoken by a speaker in view of Huffman's learning of representations of speech demonstrates building the replacement word based on representations of the speaker's voice.)
Regarding claim 54, Danieli-Bhamidipati-Feinauer-Huffman teaches all the limitations of claim 53 as laid out above. Danieli-Bhamidipati-Feinauer-Huffman, however, does not teach the limitations of claim 54.
In a similar field of endeavor, (i.e., speech recognition and profanity filtering.) Thomson teaches the machine-readable medium of claim 53, wherein the processor circuitry is to implement a neural network to maintain characteristics of a voice speaking the keyword in the speech signal with the different word. (Thomson teaches using neural networks trained on a speaker profile using speech patterns of the speaker. (i.e., the model is trained to maintain the patterns of the original speaker). Thomson at ¶ [0128]. As such, Thomson in view of Huffman's specialized text-to-speech model tuned to the speaker's vocal cords would be a neural network maintaining characteristics of the original speaker when generating new speech to replace the original speech.)
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date to combine the teachings of Danieli-Bhamidipati-Feinauer-Huffman with the teachings of Thomson to provide the limitations of claim 54. Doing so would have improved the accuracy of the ASR system as recognized by Thomson at ¶ [0129]. Further, As Thomson recognizes the functionality of profanity filtering and employing the invention of Thomson in profanity filtering it would have been obvious to combine Thomson with Danieli-Bhamidipati-Feinauer-Huffman to provide the limitations of claim 54.
Regarding claim 55, Danieli-Bhamidipati-Feinauer-Huffman in view of Thomson (hereinafter Danieli-Bhamidipati-Feinauer-Huffman-Thomson) teaches all the limitations of claim 54 as laid out above. Further, Huffman teaches the machine-readable medium of claim 54, wherein the processor circuitry is to: disentangle characteristics of the voice; (Huffman teaches using a separate independent neural network to extract information from speech that is not part mapping the voice based on characteristics such as timbre or pitch. (i.e., the characteristics of the voice, timbre and pitch, are disentangled when the neural network extracts information separately from the voice characteristics.) Huffman at ¶ [0063].)
learn representations of the speech in the audio signal independent of the source phoneme sequence; (Huffman teaches forming representations of speech using voice mappings. (i.e., learning representations of speech without using phoneme sequences, instead using voice mappings.) Huffman at ¶ [0063].)
Conclusion
THIS ACTION IS MADE FINAL. 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 CAMERON KENNETH YOUNG whose telephone number is (703)756-1527. The examiner can normally be reached Mon - Fri, 9:00 AM - 5:00 PM.
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, Andrew Flanders can be reached at 571-272-7516. 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.
/CAMERON KENNETH YOUNG/Examiner, Art Unit 2655
/ANDREW C FLANDERS/Supervisory Patent Examiner, Art Unit 2655
Prosecution Insights