Systems and methods for secure transcriptional generation

Abstract

A method, computer program product and computing system for receiving an input voice signal. A transcription of an input speech signal may be generated via an automatic speech recognition (ASR) system. One or more break points between one or more sensitive content portions and one or more non-sensitive content portions may be identified from the transcription. The input speech signal may be segmented into one or more sensitive content portions and one or more non-sensitive content portions based at least in part on the one or more segmentation points, thereby defining one or more sensitive content signals and one or more non-sensitive content signals.

Description

Background Art

[0001] Automatic clinical documentation (ACD) can be used to convert transcribed conversations (e.g., physicians, patients, and / or other participants, such as patient's family, nurses, physician assistants, etc.) into formatted (e.g., medical) reports. Physicians, scribes, etc. can review such reports, e.g., to ensure the accuracy of the reports.

[0002] However, when transcribing audio data that contains sensitive information, there is a risk of data leakage. This is especially true if the audio data is transcribed outside the firewall, such as by a contractor or quality documentation specialist (QDS). Therefore, the ability to generate accurate transcriptions may be vulnerable to the leakage of confidential and sensitive information. BRIEF DESCRIPTION OF THE DRAWINGS

[0003] Figure 1 is a schematic diagram of an automated clinical documentation computer system and transcript generation process coupled to a distributed computing network;

[0004] Figure 2 is incorporated Figure 1 Schematic diagram of a modular ACD system of an automated clinical documentation computer system;

[0005] Figure 3 yes Figure 2 A schematic diagram of a mixed media ACD device included in a modular ACD system;

[0006] Figure 4 yes Figure 1 A flowchart of one embodiment of a transcript generation process;

[0007] Figure 5 yes Figure 1 Schematic diagram of the transcriptional generation process;

[0008] Figure 6 to Figure 7 is based on Figure 1 Example transcripts of various implementations of the transcript generation process;

[0009] Figures 8 to 9 yes Figure 1 Schematic diagram of the transcriptional generation process;

[0010] Fig.10 yes Figure 1 A flowchart of one embodiment of a transcript generation process; and

[0011] Fig.11 yes Figure 1 Schematic diagram of the transcriptional generation process.

[0012] Like reference symbols in the various drawings indicate like elements. DETAILED DESCRIPTION

[0013] As described above, there is a risk of data leakage when transcribing audio data that contains sensitive information. This is particularly true if the audio data is transcribed outside a firewall and / or provided to others, such as by a contractor or quality documentation specialist (QDS). Therefore, the ability to generate accurate transcriptions may be vulnerable to the leakage of confidential and sensitive information. For example, a speech segment that is long enough to identify a speaker may be considered personally identifiable information (PII) (e.g., as defined in the General Data Protection Regulation (GDPR)) because the speaker can be identified. Additionally, the speech segment and the corresponding transcribed text may include other sensitive information (e.g., patient name, phone number, credit card number, etc.).

[0014] Thus, the present disclosure can address at least two leakage scenarios during transcription generation: disclosure of sensitive content by a labeler or an internal participant. A labeler is an external entity or system responsible for labeling or transcribing audio data. Labelers typically only see a small portion of the total audio data. However, there are typically many labelers, and ensuring their cooperation and all of their computers can be very challenging. Internal participants may scan large amounts of internal data (e.g., by looking for specific people or categories of information, such as credit card numbers). As will be discussed in more detail below, embodiments of the present disclosure provide a technical solution that is necessarily rooted in computing technology to provide secure transcription generation. Specifically, embodiments of the present disclosure can automatically (via automatic speech recognition (ASR), natural language understanding (NLU), and other speech processing systems) securely generate transcriptions without exposing sensitive content. In this way, embodiments of the present disclosure can allow manual labeling of input speech data to train a speech processing system or model without leaking sensitive content in the input speech signal.

[0015] The details of one or more embodiments will be set forth in the following drawings and the description. Other features and advantages will be apparent from the description, drawings, and claims.

[0016] System Overview:

[0017] refer to Figure 1, a transcription generation process 10 is shown. The transcription generation process 10 can be implemented as a server-side process, a client-side process, or a hybrid server-side / client-side process. For example, the transcription generation process 10 can be implemented as a pure server-side process via the transcription generation process 10s. Alternatively, the transcription generation process 10 can be implemented as a pure client-side process via one or more of the transcription generation process 10c1, the transcription generation process 10c2, the transcription generation process 10c3, and the transcription generation process 10c4. Alternatively, the transcription generation process 10 can be implemented as a hybrid server-side / client-side process via a combination of the transcription generation process 10s and one or more of the transcription generation process 10c1, the transcription generation process 10c2, the transcription generation process 10c3, and the transcription generation process 10c4.

[0018] Therefore, the transcription generation process 10 used in the present disclosure may include any combination of the transcription generation process 10s, the transcription generation process 10c1, the transcription generation process 10c2, the transcription generation process 10c3, and the transcription generation process 10c4.

[0019] The transcription generation process 10s may be a server application and may reside on and be executed by an automated clinical documentation (ACD) computer system 12, which may be connected to a network 14 (e.g., the Internet or a local area network). The ACD computer system 12 may include various components, examples of which may include, but are not limited to: a personal computer, a server computer, a series of server computers, a microcomputer, a mainframe computer, one or more network attached storage (NAS) systems, one or more storage area network (SAN) systems, one or more platform as a service (PaaS) systems, one or more infrastructure as a service (IaaS) systems, one or more software as a service (SaaS) systems, a cloud-based computing system, and a cloud-based storage platform.

[0020] As is known in the art, a SAN may include one or more of a personal computer, a server computer, a series of server computers, a minicomputer, a mainframe computer, a RAID device, and a NAS system. The various components of the ACD computer system 12 may execute one or more operating systems, examples of which may include, but are not limited to: Microsoft Windows Server 2003, ... tm ; Redhat Linux tm , Unix, or a custom operating system.

[0021] The instruction set and subroutines of the transcription generation process 10s may be stored on a storage device 16 coupled to the ACD computer system 12 and may be executed by one or more processors (not shown) and one or more memory architectures (not shown) included within the ACD computer system 12. Examples of storage device 16 may include, but are not limited to: a hard drive; a RAID device; a random access memory (RAM); a read-only memory (ROM); and all forms of flash memory storage devices.

[0022] Network 14 may be connected to one or more secondary networks (eg, network 18), examples of which may include, but are not limited to, a local area network, a wide area network, or an intranet, for example.

[0023] Various IO requests (e.g., IO requests 20) may be sent from transcription generation process 10s, transcription generation process 10c1, transcription generation process 10c2, transcription generation process 10c3, and / or transcription generation process 10c4 to ACD computer system 12. Examples of IO requests 20 may include, but are not limited to, data write requests (i.e., requests to write content to ACD computer system 12) and data read requests (i.e., requests to read content from ACD computer system 12).

[0024] The instruction sets and subroutines of the transcription generation process 10c1, the transcription generation process 10c2, the transcription generation process 10c3 and / or the transcription generation process 10c4 may be stored (respectively) on storage devices 20, 22, 24, 26 respectively coupled to the ACD client electronic devices 28, 30, 32, 34, and may be executed by one or more processors (not shown) and one or more memory architectures (not shown) respectively incorporated in the ACD client electronic devices 28, 30, 32, 34. The storage devices 20, 22, 24, 26 may include, but are not limited to: a hard disk drive; an optical drive; a RAID device; a random access memory (RAM); a read-only memory (ROM) and all forms of flash memory storage devices. Examples of ACD client electronic devices 28, 30, 32, 34 may include, but are not limited to, personal computing devices 28 (e.g., smartphones, personal digital assistants, laptops, notebook computers, and desktop computers), audio input devices 30 (e.g., handheld microphones, lavalier microphones, embedded microphones (such as microphones embedded in glasses, smartphones, tablets, and / or watches), and recording devices), display devices 32 (e.g., tablets, computer monitors, and smart TVs), machine vision input devices 34 (e.g., RGB imaging systems, infrared imaging systems, ultraviolet imaging systems, laser imaging systems, SONAR imaging systems, RADAR imaging systems, and thermal imaging systems), hybrid devices (e.g., a single device that includes the functionality of one or more of the above devices; not shown), audio rendering devices (e.g., a speaker system, a headphone system, or an earbud system; not shown), various medical devices (e.g., medical imaging devices, heart monitors, weight scales, thermometers, and blood pressure machines (not shown), and dedicated network devices (not shown).

[0025] Users 36, 38, 40, 42 may access ACD computer system 12 directly through network 14 or through auxiliary network 18. In addition, ACD computer system 12 may be connected to network 14 through auxiliary network 18, as shown by link line 44.

[0026] Various ACD client electronic devices (e.g., ACD client electronic devices 28, 30, 32, 34) may be coupled to the network 14 (or network 18) directly or indirectly. For example, the personal computing device 28 is shown as being directly coupled to the network 14 via a hard-wired network connection. Additionally, the machine vision input device 34 is shown as being directly coupled to the network 18 via a hard-wired network connection. The audio input device 30 is shown as being wirelessly coupled to the network 14 via a wireless communication channel 46 established between the audio input device 30 and a wireless access point (i.e., WAP) 48, which is shown as being directly coupled to the network 14. The WAP 48 may be, for example, an IEEE 802.11a, 802.11b, 802.11g, 802.11n, Wi-Fi, and / or Bluetooth device capable of establishing a wireless communication channel 46 between the audio input device 30 and the WAP 48. The display device 32 is shown as being wirelessly coupled to the network 14 via a wireless communication channel 50 established between the display device 32 and a WAP 52 , which is shown as being directly coupled to the network 14 .

[0027] Each ACD client electronic device (eg, ACD client electronic devices 28, 30, 32, 34) may execute an operating system, examples of which include, but are not limited to, Microsoft Windows tm 、Apple Macintosh tm 、RedhatLinux tm Or a customized operating system, wherein the combination of individual ACD client electronic devices (eg, ACD client electronic devices 28 , 30 , 32 , 34 ) and the ACD computer system 12 may form a modular ACD system 54 .

[0028] Also refer to Figure 2 , shows a simplified example embodiment of a modular ACD system 54 that is configured to automate clinical documentation. The modular ACD system 54 may include: a machine vision system 100 configured to obtain machine vision encounter information 102 related to a patient encounter; an audio recording system 104 configured to obtain audio encounter information 106 related to the patient encounter; and a computer system (e.g., ACD computer system 12) configured to receive the machine vision encounter information 102 and the audio encounter information 106 from the machine vision system 100 and the audio recording system 104 (respectively). The modular ACD system 54 may also include: a display rendering system 108 configured to render visual information 110; and an audio rendering system 112 configured to render audio information 114, wherein the ACD computer system 12 may be configured to provide the visual information 110 and the audio information 114 to the display rendering system 108 and the audio rendering system 112 (respectively).

[0029] Examples of the machine vision system 100 may include, but are not limited to: one or more ACD client electronic devices (e.g., ACD client electronic device 34, examples of which may include, but are not limited to, RGB imaging systems, infrared imaging systems, ultraviolet imaging systems, laser imaging systems, SONAR imaging systems, RADAR imaging systems, and thermal imaging systems). Examples of the audio recording system 104 may include, but are not limited to: one or more ACD client electronic devices (e.g., ACD client electronic device 30, examples of which may include, but are not limited to, handheld microphones, lavalier microphones, embedded microphones (e.g., microphones embedded in glasses, smartphones, tablets, and / or watches), and audio recording devices). Examples of the display rendering system 108 may include, but are not limited to: one or more ACD client electronic devices (e.g., ACD client electronic device 32, examples of which may include, but are not limited to, tablet computers, computer monitors, and smart TVs). Examples of the audio rendering system 112 may include, but are not limited to: one or more ACD client electronic devices (e.g., audio rendering device 116, examples of which may include, but are not limited to, speaker systems, headphone systems, and earbud systems).

[0030] As will be discussed in more detail below, the ACD computer system 12 may be configured to access one or more data sources 118 (e.g., a plurality of separate data sources 120, 122, 124, 126, 128), examples of which may include, but are not limited to, one or more of the following: a user profile data source, a voiceprint data source, a speech feature data source (e.g., for tuning an automatic speech recognition model), a faceprint data source, a humanoid data source, an utterance identifier data source, a wearable token identifier data source, an interaction identifier data source, a medical condition symptom data source, a prescription compatibility data source, a health insurance coverage data source, and a home health care data source. While in this particular example, five different examples of data sources 118 are shown, this is for illustrative purposes only and is not intended to limit the present disclosure, as other configurations are possible and are considered within the scope of the present disclosure.

[0031] As will be discussed in more detail below, the modular ACD system 54 may be configured to monitor a monitored space (e.g., monitored space 130) in a clinical environment, examples of which may include, but are not limited to, a physician's office, a medical facility, a medical institution, a medical laboratory, an urgent care facility, a medical clinic, an emergency room, an operating room, a hospital, a long-term care facility, a rehabilitation facility, a nursing home, and a hospice facility. Thus, an example of the above-mentioned patient encounter may include, but is not limited to, a patient visiting one or more of the above-mentioned clinical environments (e.g., a physician's office, a medical facility, a medical institution, a medical laboratory, an urgent care facility, a medical clinic, an emergency room, an operating room, a hospital, a long-term care facility, a rehabilitation facility, a nursing home, and a hospice facility).

[0032] When the clinical environment is larger or requires a higher level of resolution, the machine vision system 100 may include multiple discrete machine vision systems. As described above, examples of the machine vision system 100 may include, but are not limited to: one or more ACD client electronic devices (e.g., ACD client electronic devices 34, examples of which may include, but are not limited to, RGB imaging systems, infrared imaging systems, ultraviolet imaging systems, laser imaging systems, SONAR imaging systems, RADAR imaging systems, and thermal imaging systems). Therefore, the machine vision system 100 may include one or more of the following: RGB imaging systems, infrared imaging systems, ultraviolet imaging systems, laser imaging systems, SONAR imaging systems, RADAR imaging systems, and thermal imaging systems.

[0033] When the above clinical environment is larger or requires a higher level of resolution, the audio recording system 104 may include multiple discrete audio recording systems. As described above, examples of the audio recording system 104 may include, but are not limited to: one or more ACD client electronic devices (e.g., ACD client electronic devices 30, examples of which may include, but are not limited to, handheld microphones, lavalier microphones, embedded microphones (such as microphones embedded in glasses, smartphones, tablet computers and / or watches), and audio recording devices). Therefore, the audio recording system 104 may include one or more of the following: a handheld microphone, a lavalier microphone, an embedded microphone (such as a microphone embedded in glasses, smartphones, tablet computers and / or watches), and an audio recording device.

[0034] When the clinical environment is large or requires a higher level of resolution, the display rendering system 108 may include multiple discrete display rendering systems. As described above, examples of the display rendering system 108 may include, but are not limited to: one or more ACD client electronic devices (e.g., ACD client electronic devices 32, examples of which may include, but are not limited to, tablet computers, computer displays, and smart TVs). Therefore, the display rendering system 108 may include one or more of each of a tablet computer, a computer display, and a smart TV.

[0035] When the clinical environment is large or requires higher resolution, the audio rendering system 112 may include multiple discrete audio rendering systems. As described above, examples of the audio rendering system 112 may include, but are not limited to, one or more ACD client electronic devices (e.g., audio rendering device 116, examples of which may include, but are not limited to, speaker systems, headphone systems, or earplug systems). Therefore, the audio rendering system 112 may include one or more of each of the speaker system, headphone system, or earplug system.

[0036] The ACD computer system 12 may include multiple discrete computer systems. As described above, the ACD computer system 12 may include various components, examples of which may include, but are not limited to: a personal computer, a server computer, a series of server computers, a minicomputer, a mainframe computer, one or more network attached storage (NAS) systems, one or more storage area network (SAN) systems, one or more platform as a service (PaaS) systems, one or more infrastructure as a service (IaaS) systems, one or more software as a service (SaaS) systems, a cloud-based computing system, and a cloud-based storage platform. Therefore, the ACD computer system 12 may include one or more of each of the following: a personal computer, a server computer, a series of server computers, a minicomputer, a mainframe computer, one or more network attached storage (NAS) systems, one or more storage area network (SAN) systems, one or more platform as a service (PaaS) systems, one or more infrastructure as a service (IaaS) systems, one or more software as a service (SaaS) systems, a cloud-based computing system, and a cloud-based storage platform.

[0037] Also refer to Figure 3, the audio recording system 104 may include a directional microphone array 200 having a plurality of discrete microphone components. For example, the audio recording system 104 may include a plurality of discrete audio collection devices (e.g., audio collection devices 202, 204, 206, 208, 210, 212, 214, 216, 218) that may form the microphone array 200. As will be discussed in more detail below, the modular ACD system 54 may be configured to form one or more audio recording beams (e.g., audio recording beams 220, 222, 224) through the discrete audio collection devices (e.g., audio collection devices 202, 204, 206, 208, 210, 212, 214, 216, 218) included within the audio recording system 104.

[0038] For example, the modular ACD system 54 may also be configured to direct one or more audio recording beams (e.g., audio recording beams 220, 222, 224) toward one or more meeting participants (e.g., meeting participants 226, 228, 230) of the above-mentioned patient meeting. Examples of meeting participants (e.g., meeting participants 226, 228, 230) may include, but are not limited to: medical professionals (e.g., doctors, nurses, physician assistants, laboratory technicians, physical therapists, scribes (e.g., transcriptionists) and / or staff participating in patient meetings), patients (e.g., persons visiting the above-mentioned clinical environment for patient meetings), and third parties (e.g., friends of patients, relatives of patients, and / or acquaintances of patients participating in patient meetings).

[0039] Thus, the modular ACD system 54 and / or the audio recording system 104 may be configured to form an audio recording beam using one or more discrete audio collection devices (e.g., audio collection devices 202, 204, 206, 208, 210, 212, 214, 216, 218). For example, the modular ACD system 54 and / or the audio recording system 104 may be configured to form an audio recording beam 220 using the audio collection device 210 so as to be able to capture audio (e.g., speech) generated by the meeting participant 226 (when the audio collection device 210 is pointed at (i.e., toward) the meeting participant 226). Additionally, the modular ACD system 54 and / or the audio recording system 104 may be configured to form an audio recording beam 222 using the audio collection devices 204, 206 so as to be able to capture audio (e.g., speech) generated by the meeting participant 228 (when the audio collection devices 204, 206 are pointed at (i.e., toward) the meeting participant 228). Additionally, the modular ACD system 54 and / or the audio recording system 104 may be configured to form an audio recording beam 224 using the audio capture devices 212, 214 so as to capture audio (e.g., speech) produced by the meeting participant 230 (when the audio capture devices 212, 214 are pointed (i.e., toward) the meeting participant 230). Furthermore, the modular ACD system 54 and / or the audio recording system 104 may be configured to utilize zero-turn precoding to eliminate interference and / or noise between speakers.

[0040] As is known in the art, null-steering precoding is a spatial signal processing method by which a multi-antenna transmitter can null multi-user interference signals in wireless communications, wherein null-steering precoding can mitigate the effects of background noise and unknown user interference.

[0041] Specifically, null-steering precoding can be a beamforming method for narrowband signals that can compensate for the delay in receiving a signal from a particular source at different elements of an antenna array. Typically, to improve the performance of an antenna array, incoming signals can be summed and averaged, certain signals can be weighted, and signal delays can be compensated.

[0042] The machine vision system 100 and the audio recording system 104 may be separate devices (e.g. Figure 2). Additionally / alternatively, the machine vision system 100 and the audio recording system 104 can be combined into one package to form a hybrid media ACD device 232. For example, the hybrid media ACD device 232 can be configured to be mounted to structures (e.g., walls, ceilings, beams, columns) within the above-mentioned clinical environments (e.g., doctors' offices, medical facilities, medical institutions, medical laboratories, urgent care facilities, medical clinics, emergency rooms, operating rooms, hospitals, long-term care facilities, rehabilitation facilities, nursing homes, and hospice facilities), thereby allowing easy installation. In addition, when the above-mentioned clinical environment is larger or requires a higher level of resolution, the modular ACD system 54 can be configured to include multiple hybrid media ACD devices (e.g., hybrid media ACD device 232).

[0043] The modular ACD system 54 may also be configured to direct one or more audio recording beams (e.g., audio recording beams 220, 222, 224) toward one or more encounter participants (e.g., encounter participants 226, 228, 230) of a patient encounter based at least in part on the machine vision encounter information 102. As described above, the hybrid media ACD device 232 (and the machine vision system 100 / audio recording system 104 included therein) may be configured to monitor one or more encounter participants (e.g., encounter participants 226, 228, 230) of a patient encounter.

[0044] Specifically, the machine vision system 100 (as a standalone system or as a component of the hybrid media ACD device 232) can be configured to detect humanoid shapes within the above-mentioned clinical environments (e.g., doctors' offices, medical facilities, medical institutions, medical laboratories, urgent care facilities, medical clinics, emergency rooms, operating rooms, hospitals, long-term care facilities, rehabilitation facilities, nursing homes, and hospices). And, when the machine vision system 100 detects these humanoid shapes, the modular ACD system 54 and / or the audio recording system 104 can be configured to form an audio recording beam (e.g., audio recording beams 220, 222, 224) toward each detected humanoid shape (e.g., meeting participants 226, 228, 230) using one or more discrete audio capture devices (e.g., audio capture devices 202, 204, 206, 208, 210, 212, 214, 216, 218).

[0045] As described above, the ACD computer system 12 may be configured to receive machine vision meeting information 102 and audio meeting information 106 from the machine vision system 100 and the audio recording system 104 (respectively); and may be configured to provide visual information 110 and audio information 114 to the display rendering system 108 and the audio rendering system 112 (respectively). Depending on how the modular ACD system 54 (and / or the hybrid media ACD device 232) is configured, the ACD computer system 12 may be included within the hybrid media ACD device 232 or external to the hybrid media ACD device 232.

[0046] As described above, the ACD computer system 12 may execute all or part of the transcription generation process 10, wherein the instruction set and subroutines of the transcription generation process 10 (which may be stored, for example, on one or more of the storage devices 16, 20, 22, 24, 26) may be executed by the ACD computer system 12 and / or one or more of the ACD client electronic devices 28, 30, 32, 34.

[0047] Transcription generation process:

[0048] As described above, there is a risk of data leakage when transcribing audio data that contains sensitive information. This is particularly true if the audio data is transcribed outside a firewall and / or provided to others, for example, by a contractor or quality documentation specialist (QDS). Therefore, the ability to generate accurate transcriptions may be vulnerable to the leakage of confidential and sensitive information. For example, a speech segment that is long enough to identify a speaker may be considered personally identifiable information (PII) (e.g., as defined in the General Data Protection Regulation (GDPR)) because the speaker can be identified. Additionally, the speech segment and the corresponding transcribed text may contain other sensitive information (e.g., patient name, phone number, credit card number, etc.).

[0049] Therefore, the present disclosure can solve at least two leakage situations in the transcription generation process: sensitive content is leaked by a labeler or an internal participant. A labeler is an external entity or system responsible for labeling or transcribing audio data. Labelers usually only see a small part of the total audio data. However, there are usually many labelers, and it can be very challenging to ensure their cooperation and all their computers. Internal participants may scan a large amount of internal data (for example, by looking for specific people or information categories, such as credit card numbers). As will be discussed in more detail below, embodiments of the present disclosure provide a technical solution that is necessarily rooted in computing technology to provide secure transcription generation. Specifically, embodiments of the present disclosure can automatically (through automatic speech recognition (ASR), natural language understanding (NLU) and other speech processing systems) safely generate transcriptions without exposing sensitive content. In this way, embodiments of the present disclosure can allow manual labeling of input speech data to train a speech processing system or model without leaking sensitive content in the input speech signal.

[0050] At least reference Figure 4-Figure 9 , the transcription generation process 10 may receive 400 an input speech signal. A transcription of the input speech signal may be received 402. One or more sensitive content portions may be identified 404 from the transcription of the input speech signal. One or more sensitive content portions from the transcription of the input speech signal may be obfuscated 406 to define an obfuscated transcription of the input speech signal. An obfuscated speech signal may be generated 408 based at least in part on the input speech signal and the obfuscated transcription of the input speech signal.

[0051] In some embodiments consistent with the present disclosure, systems and methods for generating a novel speech "transcoder" may be provided, which has three inputs: an original speech signal, a transcription of the original speech signal, and a modified transcription in which sensitive content is obscured. As will be described in more detail below, when the transcription is not modified, the transcoder can generate an obfuscated speech signal by "transcoding" the speech to a standard speaker; but when the transcription is modified, the transcoder can create a synthesized speech that is also consistent with a standard speaker.

[0052] In some implementations, the transcription generation process 10 can receive 400 an input speech signal. For example, as described above, an audio recording system (e.g., audio recording system 104) can receive and record the input speech signal. Figure 5, the transcription generation process 10 may receive 400 an input speech signal (e.g., input speech signal 500). In one example, the input speech signal 500 may be received and recorded by an audio recording system (e.g., audio recording system 104) and / or may be a previously recorded audio input signal (e.g., an audio signal stored in a database or other data structure). In one example, assume that the input speech signal 500 relates to a medical encounter between a medical professional (e.g., participant 226) and a patient (e.g., participant 228). In this example, the medical professional (e.g., participant 226) may ask the patient (e.g., participant 228) to vocally confirm personal identification information (e.g., name, date of birth, marital status, etc.) during a physical examination. Additionally, the patient (e.g., participant 228) may describe personal health information (e.g., symptoms, medical history, etc.). Therefore, the input speech signal 500 may contain sensitive content.

[0053] In some implementations, the transcription generation process 10 may receive 402 a transcription of an input speech signal. For example, the transcription generation process 10 may provide an input speech signal (e.g., input speech signal 500) to an automatic speech recognition (ASR) system (e.g., ASR system 502) to generate a transcription (e.g., transcription 504) of the input speech signal (e.g., input speech signal 500). As is known in the art, an automatic speech recognition system may convert an input speech signal into output text. Thus, the ASR system 502 may automatically generate a transcription 504 of the input speech signal 500. As will be discussed in more detail below, the transcription 504 may include any sensitive content information recorded in the input speech signal 500.

[0054] In some embodiments, the transcription generation process 10 may identify 404 one or more sensitive content portions from the transcription of the input speech signal. Sensitive content portions may generally include any personal, private, or confidentiality-bound information fragment or type. For example, one or more sensitive content portions may include one or more of the following: personally identifiable information (PII) and protected health information (PHI). In addition to PII and PHI, sensitive content portions may also include financial information, intellectual property, trade secrets, and / or information declared private by law or regulation. Therefore, it should be understood that the transcription generation process 10 may identify 404 various types of information as sensitive content within the scope of the present disclosure.

[0055] For example, the transcript generation process 10 may utilize a sensitive content identification system (e.g., sensitive content identification system 506) to identify one or more sensitive content portions in a transcript (e.g., transcript 504). The sensitive content identification system 506 may include various known components, such as a natural language understanding (NLU) system, an artificial intelligence / machine learning model, predefined detection rules, etc., for identifying 404 one or more sensitive content portions in the transcript. The transcript generation process 10 may provide a user interface, database, and / or other data structure for examples and / or rules for identifying sensitive content in a transcript.

[0056] Also refer to Figure 6 , consider an example transcription 504 generated from an input speech signal 500 for a conversation between "Doctor Jones" and "Patient James." In this example, the transcription generation process 10 may identify 404 one or more sensitive content portions from the transcription 504. For example, the transcription generation process 10 may identify 404 the names "James" and "Jones"; the patient's full name "James Alan Alexander"; the patient's date of birth "October 12, 1987"; the patient's history of migraines; and the prescription and dosage "Migraine Ketone four times a day." While several examples of sensitive content portions have been described, it will be appreciated that these are for example purposes only, and that any number and / or type of predefined sensitive content portions may be identified within the scope of the present disclosure.

[0057] In some embodiments, the transcription generation process 10 may blur 406 one or more sensitive content parts from the transcription of the input speech signal, thereby defining a blurred transcription of the input speech signal. Blurring 406 one or more sensitive content parts from the transcription may generally include replacing, modifying and / or deleting sensitive content parts from the transcription. For example, the transcription generation process 10 may utilize various known components (such as natural language understanding (NLU) systems, artificial intelligence / machine learning models, predefined detection rules, etc.) to blur (i.e., replace and / or delete) specific parts of sensitive content. Blurring 406 one or more sensitive content parts from the transcription may include changing personally identifiable information (PII) and / or protected health information (PHI). For example, the transcription generation process 10 may include rules for replacing a specific type of sensitive content with a similar type of content. In this way, the transcription generation process 10 can blur 406 sensitive content specific to an individual associated with a specific input speech signal.

[0058] Also refer to Figure 7 , assuming that the transcript generation process 10 identifies 404 the above-mentioned sensitive content portion (e.g. Figure 6). In this example, the transcription generation process 10 may obfuscate 406 the doctor's name (e.g., replacing "Jones" with "James") and the patient's name (e.g., replacing "James Alan Alexander" with "Jason Aaron Alexander"); the patient's date of birth (e.g., replacing "October 12, 1987" with "November 12, 1997"); and / or the patient's medical history / prescription dosage information (e.g., replacing a dosage of "migraine ketone four times a day" with "migraine ketone twice a day"). The transcription generation process 10 may output these obfuscated sensitive data portions in the form of an obfuscated transcription (e.g., obfuscated transcription 508). Although several examples of obfuscating 406 sensitive content portions have been described above, it will be appreciated that within the scope of the present disclosure, the transcription generation process 10 may obfuscate 406 any combination of sensitive content portions.

[0059] In some embodiments, the transcription generation process 10 can generate 408 an ambiguous speech signal based at least in part on an input speech signal and an ambiguous transcription of the input speech signal. In some embodiments, an ambiguous speech signal can be generated using a transcoder. Conventional transcoding methods convert the properties of a signal (i.e., bit rate, encoding style, etc.) from one encoding format to a target encoding format. However, as used herein, transcoding performed by a transcoder can typically include changing the apparent "sound" and content of the input speech signal by converting speech content and / or speech signal properties. In this way, as used herein, transcoding combines two available speech processing technologies: text-to-speech (TTS) and speech conversion (e.g., hiding speaker identity information) to modify the speaker's voice and the content of the input speech signal. Conventional methods for generating secure transcriptions are unable to generate speech signals that take into account ambiguous sensitive content from speech signals or transcriptions.

[0060] Thus, the transcription generation process 10 can generate 408 fuzzy speech signals that create speech content of the modified portion of the fuzzy transcription relative to the original transcription. Example transcoder architectures / models may include Parrotron and Tacotron. As is known in the art, Parrotron is configured to perform speech-to-speech conversion, while Tacotron is configured to perform text-to-speech conversion. Each of these example transcoder architectures / models includes an encoder-decoder architecture that maps input (e.g., text in the case of Tacotron, and a logarithmic Mel frequency representation of a speech signal in the case of Parrotron) to encoder states and decoders to convert the input to speech. Although examples of Parrotron and Tacotron have been provided for parts of the transcoder architecture / model, it is understood that any transcoder architecture / model may be used within the scope of the present disclosure. Additionally / alternatively, a text-to-speech (TTS) system configured to convert a transcription of an input speech signal into a fuzzy speech signal may be used to generate a fuzzy speech signal.

[0061] Generating 408 an obfuscated speech signal based at least in part on the input speech signal and the obfuscated transcription of the input speech signal may include identifying 410 one or more sensitive content portions obfuscated in the obfuscated transcription of the input speech signal, thereby defining the one or more obfuscated sensitive content portions. For example, when obfuscating one or more sensitive content portions, the transcription generation process 10 may maintain a database, record, or log of the obfuscated portions. In this way, the transcription generation process 10 may identify each sensitive content portion to synthesize the obfuscated speech signal.

[0062] Generating 408 an obfuscated speech signal based at least in part on an input speech signal and an obfuscated transcription of the input speech signal may include synthesizing 412 an obfuscated sensitive content portion to generate an obfuscated speech signal. For example, the transcription generation process 10 may identify an obfuscated sensitive content portion in the obfuscated transcription. Thus, the transcription generation process 10 may synthesize 412 the obfuscated sensitive content portion to generate an obfuscated speech signal. Synthesizing 412 a corresponding portion of the obfuscated transcription may typically include generating a speech output (e.g., via a speech-to-text function of a transcoder) based on the corresponding portion of the obfuscated transcription. In this way, a subsequent listener of the obfuscated speech signal may hear a consistent speech signal, even though portions of the original speech signal are obfuscated.

[0063] Generating 408 an ambiguous speech signal based at least in part on an input speech signal and an ambiguous transcription of the input speech signal may include comparing 414 a portion of the transcription of the input speech signal to a corresponding portion of the ambiguous transcription of the input speech signal. For example, the transcription generation process 10 may configure a transcoder (e.g., transcoder 510) to modify an input speech signal (e.g., input speech signal 500) to generate an ambiguous speech signal of an unambiguous portion and synthesize a modified portion of the ambiguous transcription to generate an ambiguous speech signal of the modified portion. Figure 5 A transcoder (e.g., transcoder 510) is shown, but it should be understood that this is for example purposes only and that various systems (e.g., text-to-speech systems) may be used to generate the fuzzy speech signal within the scope of the present disclosure. The transcription generation process 10 may compare each portion of the transcription of the input speech signal (e.g., each character, word, sentence, paragraph, etc.) to the corresponding portion of the fuzzy transcription of the input speech signal 414. Although examples of various granularities for comparing transcriptions have been described, it should be understood that the transcription generation process 10 may process and compare transcriptions at any granularity order within the scope of the present disclosure.

[0064] Comparing 414 the transcribed portion of the input speech signal to the corresponding portion of the ambiguous transcription of the input speech signal may include determining whether the transcribed portion of the input speech signal is identical to the corresponding portion of the ambiguous transcription of the input speech signal. Figure 6-Figure 7 , the transcript generation process 10 can iteratively and sequentially compare each portion of the transcript 504 to each corresponding portion of the fuzzy transcript 508 to determine whether each portion is identical. For example, the transcript generation process 10 can Figure 6 Each word, sentence, and / or paragraph of the transcription 504 shown in Figure 7 Each word, sentence, and / or paragraph of the fuzzy transcription 508 shown in is compared 414 .

[0065] In response to determining that the portion of the transcription of the input speech signal is different from the corresponding portion of the ambiguous transcription of the input speech signal, the transcription generation process 10 can synthesize 416 the corresponding portion of the ambiguous transcription of the input speech signal to generate an ambiguous speech signal. For example, the transcription generation process 10 can identify the portion of the ambiguous transcription 508 that does not match the corresponding portion of the transcription 504. Therefore, the transcription generation process 10 can synthesize 416 the corresponding portion of the ambiguous transcription into the portion of the ambiguous speech signal through the transcoder 510. Synthesizing 416 the corresponding portion of the ambiguous transcription can generally include generating a speech output based on the corresponding portion of the ambiguous transcription (e.g., through a speech-to-text function of the transcoder). In this way, a subsequent listener of the ambiguous speech signal can hear a consistent speech signal even though a portion of the original speech signal is blurred.

[0066] In response to determining that the portion of the transcription of the input speech signal is the same as the corresponding portion of the fuzzy transcription of the input speech signal, the transcription generation process 10 can use 418 the corresponding portion of the input speech signal to generate a fuzzy speech signal. For example, the transcription generation process 10 can use 418 the speech-to-speech function of the transcoder 510 to convert the portion of the input speech signal into the same output type as the unfuzzy portion and the fuzzy portion. In this way, although the portions of the input speech signal are blurred in the fuzzy transcription, the fuzzy speech signal (e.g., the fuzzy speech signal 512) can be auditorily consistent. As will be discussed in more detail below, by having consistent auditory characteristics in the fuzzy speech signal, the speech processing system (e.g., the speech processing system 514) can be more accurately trained to process the input speech of the target output text. In some embodiments, as opposed to using the original input speech signal, the transcription of the input speech signal can be processed using a text-to-speech system so that the final version of the input speech signal will not be based on the original input signal. In this way, the original input speech signal can be prevented from being used in a manner that may expose sensitive content.

[0067] Generating 408 the obfuscated speech signal based at least in part on the input speech signal, the transcription of the input speech signal, and the obfuscated transcription of the input speech signal may include modifying 420 one or more attributes of the input speech signal to mask personally identifiable information within the input speech signal. For example, as described above, one of the regulatory challenges associated with speech signals is the risk of being able to identify the identity of a speaker based on a sufficiently long recording of the speaker's voice. For example, assume that the input speech signal 500 includes a conversation between a patient and a medical professional, but the medical professional never says his or her name. Despite the lack of personally identifiable information in the content of the conversation, the speaker's voice characteristics may allow the speaker to be identified.

[0068] Therefore, the transcription generation process 10 can perform various speech signal modifications on the input speech signal through a transcoder (e.g., transcoder 510) to reduce the possibility that the speaker's voice can be personally identified. For example, the transcription generation process 10 can modify 420 the gain, noise, reverberation, rhythm, speed or other properties of the input speech signal (or a specific part (i.e., time period, target frequency, etc.)) to avoid retaining personally identifiable information. In this way, the transcription generation process 10 can blur content-based sensitive information and spectrum-sensitive information from the input speech signal (e.g., input speech signal 500). Additionally / alternatively, as described above, the transcription of the input speech signal can be processed using a text-to-speech system so that the final version of the input speech signal will not be based on the original input signal. In this way, the original input speech signal can be avoided from being used in a manner that may expose sensitive content. Therefore, the transcription generation process 10 can perform various speech signal modifications on the revised version of the input speech signal through a transcoder (e.g., transcoder 510) to reduce the possibility that the speaker's voice can be personally identified.

[0069] In some implementations, the transcription generation process 10 may receive 422 a manually generated transcription of an ambiguous speech signal. For example, the transcription generation process 10 may provide the ambiguous speech signal (e.g., the ambiguous speech signal 512) to a transcriptionist (e.g., transcriptionist 516) to generate a manually generated transcription of the ambiguous speech signal. As described above, the transcription generation process 10 may utilize the collected speech signal (e.g., the input speech signal 500) to improve the accuracy of the speech processing system (e.g., speech processing system 514) by training the speech processing system using labeled data (i.e., in the case of speech-to-text processing, the input speech data and the corresponding text data). Thus, the transcription generation process 10 may provide the ambiguous speech signal (e.g., the ambiguous speech signal 512) to a transcriptionist (e.g., transcriptionist 516) to generate a transcription of the ambiguous speech signal. The transcriptionist (e.g., transcriptionist 516) may typically include a person (e.g., user 518) using a software system and / or a hybrid approach including a speech processing system (e.g., another ASR system 520). Thus, it will be appreciated that the transcribers 516 may include any group of users and / or software models configured to process and label the ambiguous speech signal 512 .

[0070] As described above, by generating the obfuscated speech signal 512, the transcription generation process 10 can obfuscate 406 and / or modify 420 the input speech signal 500, thereby eliminating or reducing the possibility of sharing sensitive information with the transcriptionist 516. Figure 5As shown, the transcription generation process 10 can receive 422 a transcription of the ambiguous speech signal (e.g., transcription 522) from the transcriptionist 516. As described above, the transcription 522 can include content of the labeled input speech signal 512 for use in training a speech processing system (e.g., speech processing system 514).

[0071] In some embodiments, the transcription generation process 10 may train 424 a speech processing model based at least in part on a manually generated transcription of an ambiguous speech signal. As described above, training 424 a speech processing system (e.g., speech processing system 514) may include providing labeled data and input speech data for the speech processing system to "learn". For example, as is known in the art, machine learning generally includes training an algorithm or a combination of algorithms to recognize certain types of patterns. Machine learning methods can generally be divided into three categories, depending on the nature of the available signals: supervised learning, unsupervised learning, and reinforcement learning. As is known in the art, supervised learning may include presenting example inputs and their expected outputs given by a "teacher" to a computing device, with the goal of learning general rules that map inputs to outputs. For unsupervised learning, no labels are provided to the learning algorithm, allowing it to find structure in its input on its own. Unsupervised learning can be a goal in itself (discovering hidden patterns in data) or a means to achieve a goal (feature learning). As is known in the art, reinforcement learning may generally include a computing device interacting in a dynamic environment in which it must perform a certain goal (such as driving a vehicle or playing a game with an opponent). As it navigates in the problem space, the program is provided with feedback similar to a reward, which it tries to maximize. Although three examples of machine learning methods have been provided, it will be appreciated that other machine learning methods are possible within the scope of the present disclosure.

[0072] Training 424 a speech processing model based at least in part on a manually generated transcription of an ambiguous speech signal may include training 426 a speech processing model based at least in part on a manually generated transcription of an ambiguous speech signal and an ambiguous speech signal. In this example, the transcription generation process 10 may train 426 a speech processing system (e.g., speech processing system 514) using a combination of an ambiguous speech signal (e.g., ambiguous speech signal 512) and a manually generated transcription of the ambiguous speech signal (e.g., manually generated transcription 522). Thus, the manually generated transcription 522 may serve as labeled data for the ambiguous speech signal. This approach may also allow the original input speech signal (e.g., input speech signal 500) to be discarded. In this way, the transcription generation process 10 may utilize an input speech signal that originally contained a portion of sensitive content for training data without disclosing the sensitive content to an external party. Additionally, by discarding the original input speech signal, data privacy requirements may be reduced or avoided entirely.

[0073] Training 424 the speech processing model based at least in part on the manually generated transcription of the obfuscated speech signal may include aligning 428 the manually generated transcription of the obfuscated speech signal with the transcription of the input speech signal; and incorporating 430 one or more sensitive content portions into the manually generated transcription of the obfuscated speech signal, thereby defining a modified manually generated transcription. For example and also with reference to Figure 8 , it may be beneficial to use the content of the original input speech signal to mark the manually generated transcription of the ambiguous speech signal. Therefore, the transcription generation process 10 can align 428 the manually generated transcription of the ambiguous speech signal (e.g., the manually generated transcription 522) with the transcription of the input speech signal (e.g., the transcription 504) through the aligner 800. Aligning 428 the manually generated transcription of the ambiguous speech signal with the transcription can generally include identifying corresponding portions from each transcription to synchronize the transcriptions. In this way, the transcription generation process 10 can generate a hybrid automatic and manual transcription of the input speech signal.

[0074] In the event that the manually generated transcription of the ambiguous speech signal (e.g., manually generated transcription 522) is aligned with the transcription of the input speech signal (e.g., transcription 504), the transcription generation process 10 can incorporate 430 one or more sensitive content portions into the manually generated transcription of the ambiguous speech signal (e.g., manually generated transcription 522) to generate a modified manually generated transcription (e.g., modified manually generated transcription 802). In this way, the transcription generation process 10 can reintroduce the sensitive content portions into the manually generated transcription to allow the speech processing system (e.g., speech processing system 514) to be trained using the original input speech signal (e.g., input speech signal 500) and the modified manually generated transcription (e.g., modified manually generated transcription 802). Therefore, the transcription generation process 10 can provide the modified manually generated transcription (e.g., modified manually generated transcription 802) and the input speech signal (e.g., input speech signal 500) as training data to the speech processing system 514.

[0075] In some embodiments, the transcript generation process 10 may allow for the generation of reports (i.e., a medical report from a medical encounter transcript and / or various reports of a particular encounter transcript) while limiting exposure of sensitive content from other users (e.g., scribes). Fig. 9 and with Figure 5By comparison, the transcriptionist 516 may generate an ambiguous report (e.g., ambiguous report 900) from an ambiguous speech signal (e.g., ambiguous speech signal 512). However, since the ambiguous report 900 does not include sensitive content, the transcription process 10 may utilize a mapping between the ASR input and the output of the sensitive content identification system 506 to map the sensitive content to ambiguous data in the ambiguous report (e.g., ambiguous report 900). The mapping 902 may represent one or more mappings generated between the input speech signal 500 and the output of the sensitive content identification system 506. The mapping 902 may be a user-defined predefined mapping and / or may be automatically and dynamically defined by the transcription generation process 10 using one or more machine learning models based on the ASR input and the output of the sensitive content identification system 506. In this way, the transcription generation process 10 may utilize the ASR input (e.g., the input speech signal 500), the ambiguous report 900, and the mapping 902 to generate a complete report (e.g., complete report 904) including sensitive content.

[0076] At least reference Figure 10-11 , a transcription generation process 10 may receive 1000 an input speech signal. A transcription of the input speech signal may be generated 1002 by an automatic speech recognition (ASR) system. One or more segmentation points between one or more sensitive content portions and one or more non-sensitive content portions from the transcription may be identified 1004. The input speech signal may be segmented 1006 into one or more sensitive content portions and one or more non-sensitive content portions based at least in part on the one or more segmentation points, thereby defining one or more sensitive content signals and one or more non-sensitive content signals.

[0077] In some embodiments consistent with the present disclosure, systems and methods may be provided for generating transcriptions without exposing sensitive content in an input voice signal. For example, as will be discussed in more detail below, ASR and NLU may be used to identify sensitive content (e.g., patient name and medical record number) from an input voice signal. The voice signal may be segmented into different audios for transcription, which are sent to different transcribers. One audio may include the sensitive content (and a few seconds of "buffering" before and after each segment), and another audio may include the rest of the original input voice signal. These audios may be sent to different transcribers so that no single transcriber can leak the entire conversation. This measure may result in less harmful leakage of sensitive content and may make any stored voice data less attractive to potential leakers.

[0078] In some implementations, the transcription generation process 10 can receive 1000 an input speech signal. For example, as described above, an audio recording system (e.g., audio recording system 104) can receive and record the input speech signal. Fig.11, the transcription generation process 10 may receive 1000 an input speech signal (e.g., input speech signal 1100). In one example, the input speech signal 1000 may be received and recorded by an audio recording system (e.g., audio recording system 104) and / or may be a previously recorded audio input signal (e.g., an audio signal stored in a database or other data structure). In one example, assume that the input speech signal 1100 relates to a medical encounter between a medical professional (e.g., participant 226) and a patient (e.g., participant 228). In this example, the medical professional (e.g., participant 226) may ask the patient (e.g., participant 228) to confirm personal identification information (e.g., name, date of birth, marital status, etc.) in an audible manner during a physical examination. Additionally, the patient (e.g., participant 228) may describe personal health information (e.g., symptoms, medical history, etc.). Therefore, the input speech signal 1100 may contain sensitive content information.

[0079] The transcription generation process 10 can generate 1002 a transcription of an input speech signal via an automatic speech recognition (ASR) system. For example, the transcription generation process 10 can provide an input speech signal (e.g., input speech signal 1100) to an automatic speech recognition (ASR) system (e.g., ASR system 1102) to generate a transcription (e.g., transcription 1104) of the input speech signal (e.g., input speech signal 1100). As is known in the art, an automatic speech recognition system can convert an input speech signal into output text. Thus, the ASR system 1002 can automatically generate a transcription 1104 of the input speech signal 1100. As will be discussed in more detail below, the transcription 1104 can include any sensitive content information recorded in the input speech signal 1100.

[0080] In some embodiments, the transcription generation process 10 may identify 1004 one or more segmentation points between one or more sensitive content parts and one or more non-sensitive content parts from the transcription. For example, as described above, the transcription generation process 10 may identify one or more sensitive content parts from the transcription of the input speech signal. The sensitive content part may generally include any personal, private, or confidentiality-constrained information fragment or information type. For example, one or more sensitive content parts may include one or more of the following: personally identifiable information (PII) and protected health information (PHI). In addition to PII and PHI, the sensitive content part may also include financial information, intellectual property, trade secrets, and / or information declared private by law or regulation. Therefore, it should be understood that the transcription generation process 10 can identify various types of information as sensitive content.

[0081] For example, the transcript generation process 10 may utilize a sensitive content identification system (e.g., sensitive content identification system 1106) to identify one or more sensitive content portions in a transcript (e.g., transcript 1104). The sensitive content identification system 1106 may include various known components, such as a natural language understanding (NLU) system, an artificial intelligence / machine learning model, predefined detection rules, etc., for identifying one or more sensitive content portions in the transcript 1004. The transcript generation process 10 may provide a user interface, database, and / or other data structure for examples and / or rules for identifying sensitive content in a transcript.

[0082] Identifying 1004 one or more segmentation points between one or more sensitive content portions and one or more non-sensitive content portions from the transcript may include identifying one or more sensitive content portions within the transcript. Figure 6 In the example transcription of , the transcription generation process 10 can identify various sensitive content portions (e.g., "James"; "Doctor Jones"; "James Alan Alexander"; "October 12, 1987"; "migraine"; "migraine ketone four times a day"). Therefore, the transcription generation process 10 can identify one or more segmentation points in the input speech signal to isolate the audio portion of the sensitive content portion. For example, the transcription generation process 10 can identify timestamps before and after each sensitive content portion in the input speech signal corresponding to the sensitive content portion, thereby defining one or more segmentation points (e.g., one or more segmentation points 1108).

[0083] In some embodiments, one or more segmentation points may include a predefined amount of audio before and after one or more sensitive content parts. For example, assume that the sensitive content part starts at, for example, 187 seconds in the input voice signal and reaches, for example, 188.5 seconds. In this example, the transcription generation process 10 may define a segmentation point at, for example, 187 seconds and another segmentation point at, for example, 188.5 seconds. Therefore, as will be discussed in more detail below, the transcription generation process 10 may provide 0 to 187 seconds and 188.5 seconds to the end of the input voice signal to the non-sensitive content transcriptionist, while providing, for example, 186 seconds to, for example, 189.5 seconds to the sensitive content transcriptionist. In this way, the non-sensitive content transcriptionist will not receive sensitive content, and the sensitive content transcriptionist may receive a portion of the non-sensitive context (e.g., a predefined amount of audio) to assist in its transcription. The predefined amount of audio before and / or after one or more sensitive content parts may be a default amount, a user-defined amount, and / or automatically defined by the transcription generation process 10. In some embodiments, a predefined amount of audio before and / or after one or more sensitive content portions may be provided in one or more non-sensitive content portions (i.e., may be copied to appear in both the non-sensitive content portions and the sensitive content portions) rather than being missing or deleted from one or more non-sensitive content portions.

[0084] In some embodiments, identifying 1004 one or more segmentation points between one or more sensitive content portions and one or more non-sensitive content portions from the transcription may include identifying one or more sensitive content portion types within the transcription. For example, the transcription generation process 10 may identify 1004 sensitive content portions that belong to a unique type of sensitive content. In one example, the transcription generation process 10 may identify a person's legal name; a person's medical information; financial information (i.e., bank account information); or other types of sensitive content. For example, a natural language understanding (NLU) system, an artificial intelligence / machine learning model, predefined detection rules, etc. may be used to identify specific sensitive content portion categories. In some embodiments, one or more sensitive content portion categories may be predefined, default categories, user-defined (e.g., via a user interface), and / or learned by a machine learning system algorithm. In this way, the transcription generation process 10 may identify sensitive content of different sensitive content types, respectively.

[0085] As will be discussed in more detail below, one or more segmentation points (e.g., one or more segmentation points 1108) may be used to identify sensitive audio portions and non-sensitive audio portions within an input speech signal (e.g., input speech signal 1100). Although one example of a single category of sensitive audio portions has been described, it will be appreciated that within the scope of the present disclosure, the transcription generation process 10 may identify 1004 any number of segmentation point types for segmenting the input speech signal.

[0086] The transcription generation process 10 may segment 1006 the input speech signal into one or more sensitive content parts and one or more non-sensitive content parts based at least in part on one or more segmentation points, thereby defining one or more sensitive content signals and one or more non-sensitive content signals. For example, the transcription generation process 10 may utilize a segmentation system (e.g., segmentation system 1110) to process one or more segmentation points (e.g., one or more segmentation points 1108) to segment 1006 the input speech signal (e.g., input speech signal 1100) into one or more sensitive content signals (e.g., represented by sensitive content signal 1112) including one or more sensitive content parts and one or more non-sensitive content signals (e.g., represented by non-sensitive content signal 1114) including one or more non-sensitive content parts. The segmentation system 1110 may include any known (multiple) software modules that are configured to segment the input audio content into multiple segments based on input constraints. It should be understood that the transcription generation process 10 may segment the input speech signal into any number of parts 1006 within the scope of the present disclosure. For example, various portions may be defined for different levels of content sensitivity (eg, low sensitivity, medium sensitivity, and / or high sensitivity).

[0087] Continuing with the above example, assume that the sensitive content portion (e.g., sensitive content signal 1112) starts at, for example, 187 seconds within the input speech signal 1100 and reaches, for example, 188.5 seconds. In this example, the transcription generation process 10 may define a segmentation point (e.g., one or more segmentation points 1108) at, for example, 187 seconds, and another segmentation point at, for example, 188.5 seconds. The sensitive content signal 1112 may include a portion of the input speech signal 1100 corresponding to one or more sensitive content portions identified in the transcription 1104, while the non-sensitive content speech signal 1114 may include the remainder of the input speech signal 1100. Therefore, the sensitive content signal 1112 may include primarily sensitive content without most of the context of the input speech signal 1100, while the non-sensitive content speech signal 1114 may include only the non-sensitive content portion of the input speech signal 1100. In this way, the transcription generation process 10 may isolate sensitive content from non-sensitive content. Although an example of segmenting 1006 the input speech signal 1100 into, for example, two speech signals has been provided, it should be understood that the transcription generation process 10 may segment the input speech signal 1100 into any number of signals.

[0088] For example, the transcription generation process 10 can segment the input speech signal 1100 into a plurality of speech signals based at least in part on the type of sensitive content identified in the transcription 1104. In one example, the transcription generation process 10 can segment the input speech signal 1100 into one speech signal for personally identifiable information (PII), one speech signal for protected health information (PHI), and another speech signal for all remaining content of the input speech signal 1100. Thus, it should be appreciated that the transcription generation process 10 can segment the input speech signal into any number of speech signals based at least in part on one or more segmentation points.

[0089] The transcription generation process 10 may modify 908 one or more properties of the input speech signal to mask personally identifiable information in the input speech signal. For example, as described above, one of the regulatory challenges associated with speech signals is the risk of being able to identify a speaker based on a sufficiently long recording of the speaker's voice. For example, assume that the input speech signal 1100 includes a conversation between a patient and a medical professional, but the medical professional never says his or her name. Despite the lack of personally identifiable information in the content of the conversation, the speaker's voice characteristics may allow the speaker to be identified.

[0090] Therefore, the transcription generation process 10 may perform various speech signal modifications on the input speech signal to reduce the possibility that the speaker's voice can be personally identified. For example, the transcription generation process 10 may modify 1008 the gain, noise, reverberation, rhythm, speed or other properties of the input speech signal (or a specific part (i.e., time period, target frequency, etc.)) to avoid retaining personally identifiable information. In some embodiments, the transcription generation process 10 may utilize a voice style conversion (VST) method (i.e., an audio-to-audio conversion method that accepts the input of a speech segment and a target speaker embedding and outputs audio that sounds like the target speaker's voice). Therefore, the transcription generation process 10 may modify 1008 the input speech signal by performing VST on a standard speaker, which will allow the speech segment to retain PII from the speaker's "voice" perspective, and this can be done with other processing (such as modifying pitch, etc.). In this way, the transcription generation process 10 can obscure content-based sensitive information and spectrum-sensitive information from the input speech signal (e.g., input speech signal 1000) while still sounding like the speaker's voice.

[0091] In some implementations, the transcription generation process 10 may provide 1010 the sensitive content signal to a first transcriptionist. For example, the transcription generation process 10 may provide the sensitive content signal (e.g., sensitive content signal 1112) to a transcriptionist (e.g., transcriptionist 1116) to generate a manually generated transcription of the sensitive content signal. As described above, the transcription generation process 10 may utilize the collected speech signal (e.g., input speech signal 1100) to improve the accuracy of the speech processing system (e.g., speech processing system 514) by training the speech processing system using labeled data (i.e., in the case of speech-to-text processing, the input speech data and the corresponding text data).

[0092] Thus, the transcription generation process 10 may provide 1010 a sensitive content signal (e.g., sensitive content signal 1112) to a transcriber (e.g., transcriber 1116) to generate a transcription of the sensitive content signal. The transcriber (e.g., transcriber 1116) may typically include a person (e.g., user 1118) utilizing a software system and / or a hybrid approach including a speech processing system (e.g., another ASR system 1120). Thus, it should be understood that the transcriber 1116 may include any group of users and / or software models configured to process and mark the sensitive content signal 1112. As described above, by generating the sensitive content signal 1112, the transcription generation process 10 may isolate the sensitive content of the input speech signal 1100, thereby eliminating or reducing the possibility of leaking the entirety of the input speech signal 1100. As described above, Fig.11 As shown, the transcription generation process 10 may receive 1014 a manually generated transcription of a sensitive content signal from a transcriptionist 1116, thereby defining a sensitive content transcription (e.g., sensitive content transcription 1122). Although the transcriptionist 1116 is described as a single transcriptionist, it should be understood that this is for example purposes only, and providing 1010 a sensitive content signal (e.g., sensitive content signal 1112) to a transcriptionist (e.g., transcriptionist 1116) to generate a transcription of the sensitive content signal may include providing the sensitive content signal to one or more transcriptionists. In some embodiments, the first transcriptionist or the first group of transcriptionists may include a group of trusted and / or closely supervised transcriptionists for receiving more sensitive content.

[0093] Providing 1010 the one or more sensitive content signals to the first transcriptionist may include providing 1012 each corresponding sensitive content signal of the one or more sensitive content signals to different transcriptionists in the plurality of transcriptionists. For example, as described above, assume that the transcription generation process 10 identifies multiple types of sensitive content. The transcription generation process 10 may provide 1012 these corresponding sensitive content signals to different transcriptionists. For example, one transcriptionist may receive personal medical information, while another transcriptionist may receive financial information. In this way, each transcriptionist may only have access to limited portions or types of sensitive content.

[0094] In some embodiments, the transcription generation process 10 may provide 1014 the non-sensitive content signal to a second transcriptionist. For example, the transcription generation process 10 may provide the non-sensitive content signal (e.g., non-sensitive content signal 1114) to a transcriptionist (e.g., transcriptionist 1124) to generate a manually generated transcription of the non-sensitive content signal. As described above, the transcriptionist (e.g., transcriptionist 1124) may generally include a person (e.g., user 1126) utilizing a software system and / or a hybrid approach including a speech processing system (e.g., another ASR system 1128). Therefore, it should be recognized that the transcriptionist 1124 may include any group of users and / or software models configured to process and mark the non-sensitive content signal 1114. In some embodiments, the second transcriptionist 1124 may include a transcriptionist having a different set of authorizations or permissions to access and transcribe sensitive material than the first transcriptionist (e.g., first transcriptionist 1116). For example, a first transcriptionist (e.g., transcriptionist 1116) may include the most trusted and / or most closely supervised transcriptionist for receiving more sensitive content, while non-sensitive content may be transcribed by a second transcriptionist (e.g., transcriptionist 1124) who is less supervised and / or in a less secure environment (e.g., a contractor working from home). In this way, the distinction between a first transcriptionist receiving sensitive content and a second transcriptionist receiving non-sensitive content may be based at least in part on the degree of data security and the different trust levels or data security training of the transcriptionists. As described above, by generating the non-sensitive content signal 1114, the transcription generation process 10 can remove sensitive content from the input speech signal 1100, thereby eliminating or reducing the possibility of leaking the entirety of the input speech signal 1100. Fig.11 As shown, the transcription generation process 10 may receive 1018 from the transcriptionist 1124 a manually generated transcription of the non-sensitive content signal from the second transcriptionist, thereby defining a non-sensitive content transcription (eg, non-sensitive content transcription 1130 ).

[0095] Although the above example includes two transcriptionists, it is understood that any number of transcriptionists can receive the speech signal and generate a transcription of its corresponding portion. For example, the transcription generation process 10 can provide two speech signals (e.g., sensitive content signal 1112 and non-sensitive content signal 1114) to the same transcriptionist (e.g., transcriptionist 1116) at different times. In this example, the transcription generation process 10 can provide the sensitive content signal 1112 to the transcriptionist 1116. After a predefined amount of time, the transcription generation process 10 can provide the non-sensitive content signal 1114 to the transcriptionist 1116. In this way, the transcriptionist 1116 can generate a transcription of the two signals without knowing their connection. Therefore, the transcription generation process 10 can reduce the number of transcriptionists required to transcribe the sensitive content in the input speech signal by providing the speech signals separately over time. In some embodiments, the amount of time defined between providing each speech signal can be user-defined (e.g., through a user interface), can be automatically defined by the transcription generation process 10, and / or can be a default value (e.g., three days). Therefore, it should be understood that the transcription generation process 10 may provide the segmented speech signal to the same transcriber after any period of time within the scope of the present disclosure.

[0096] Continuing with the above example, the transcription generation process 10 may provide the non-sensitive content transcriber (e.g., transcriber 1124) with 0 to 187 seconds and 188.5 seconds before the end of the input speech signal (e.g., non-sensitive content signal 1114), while providing the sensitive content transcriber (e.g., transcriber 1116) with, for example, 186 seconds to, for example, 189.5 seconds (e.g., sensitive content signal 1112). In this way, the non-sensitive content transcriber (e.g., transcriber 1124) will not receive the sensitive content (e.g., non-sensitive content signal 1114), while the sensitive content transcriber (e.g., transcriber 1116) may receive a portion of the non-sensitive context (e.g., a predefined amount of audio) to assist in its transcription.

[0097] In some embodiments, the transcription generation process 10 can combine 1020 the sensitive content transcription and the non-sensitive content transcription to define a combined transcription. For example, the sensitive content transcription 1122 can include sensitive content from the input speech signal 1100, while the non-sensitive content transcription 1130 can include all other portions of the input speech signal 1100. Thus, the transcription generation process 10 can combine 1020 the sensitive content transcription 1122 and the non-sensitive content transcription 1130 using a combination system 1132 to form a combined transcription (e.g., a combined transcription 1134) representing all of the content (i.e., sensitive and non-sensitive portions) of the input speech signal 1100. The combination system 1132 can be any known software module(s) configured to combine separate transcriptions using a reference audio input (e.g., the input speech signal 1100). In some embodiments and as Fig.11 As shown, combining system 1132 may receive input for synchronizing sensitive content transcription 1122 and non-sensitive content transcription 1130 to form a combined transcription (eg, combined transcription 1134 ) representing the content (ie, sensitive and non-sensitive portions) of input speech signal 1100 .

[0098] For example, the transcription generation process 10 may provide a synchronization / segmentation signal (in Fig.11 1132). In this way, the transcription generation process 10 can utilize the segmentation signal or information from the segmentation system 1110 to synchronize the sensitive content transcription 1122 and the non-sensitive content transcription 1130 when generating the combined transcription 1134. Although the segmentation signal / information is shown as a direct link between the segmentation system 1110 and the combining system 1132, it can be understood that the transcription generation process 10 can provide the segmentation signal / information from the segmentation system 1110 through one or more intermediate systems or processes. Additionally and / or alternatively, the transcription generation process 10 can provide information embedded in one or more of the sensitive content transcription 1122 and the non-sensitive content transcription 1130 to the combining system 1132 to allow the combining system 1132 to synchronize the sensitive content transcription 1122 and the non-sensitive content transcription 1130.

[0099] The transcription generation process 10 may train 1022 a speech processing model based at least in part on the combined transcription and the input speech signal. For example, the transcription generation process 10 may train 1022 a speech processing system (e.g., speech processing system 514) using a combination of an input speech signal (e.g., ambiguous speech signal 1100) and a combined transcription (e.g., combined transcription 1134). Thus, the combined transcription 1134 may serve as labeled data for the input speech signal. This approach may allow for labeling of input speech signals (e.g., input speech signal 1100) while reducing the likelihood of exposing all input speech signals to a particular transcriptionist. In this way, the transcription generation process 10 may utilize input speech signals that initially contain portions of sensitive content to train data without disclosing the sensitive content or all of the content to external parties.

[0100] In some implementations, the transcription generation process 10 can selectively remove sensitive content from the training of the speech processing system. For example, some or all of the sensitive content can be identified as described above, but the transcription generation process 10 can process one or more sensitive content signals (e.g., sensitive content signal 1112) instead of providing the sensitive content for transcription. In this way, the speech processing system (e.g., speech processing system 514) can be trained using only non-sensitive content from the input speech signal.

[0101] General:

[0102] Those skilled in the art will appreciate that the present disclosure may be embodied as a method, system, or computer program product. Thus, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment (including firmware, resident software, microcode, etc.), or an embodiment combining software and hardware aspects, which may generally be referred to herein as a "circuit," "module," or "system." In addition, the present disclosure may take the form of a computer program product on a computer-usable storage medium, wherein the computer-usable storage medium has a computer-usable program code embodied in the medium.

[0103] Any suitable computer-usable or computer-readable medium may be used. A computer-usable or computer-readable medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, device, apparatus or propagation medium. More specific examples of computer-readable media (a non-exhaustive list) may include the following: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a transmission medium (such as a transmission medium supporting the Internet or an intranet), or a magnetic storage device. A computer-usable or computer-readable medium may also be other suitable media for paper or printed programs, as the program may be captured electronically, such as by optical scanning of paper or other media, and then compiled, interpreted or otherwise processed (if necessary) in an appropriate manner and then stored in a computer memory. In the context of this document, a computer-usable or computer-readable medium may be any medium that can contain, store, communicate, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The computer usable medium may include a propagated data signal in which the computer usable program code is embodied, whether in baseband or as part of a carrier wave. The computer usable program code may be transmitted using any appropriate medium, including but not limited to the Internet, wireline, optical fiber cable, RF, etc.

[0104] The computer program code for performing the operations of the present disclosure can be written in an object-oriented programming language, such as Java, Smalltalk, C++, etc. However, the computer program code for performing the operations of the present disclosure can also be written in a conventional program programming language, such as the "C" programming language or a similar programming language. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer, partially on a remote computer, or completely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer via a local area network / wide area network / internet (e.g., network 14).

[0105] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each frame in the flowchart and / or block diagram and the combination of frames in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer / special-purpose computer / other programmable data processing device, so that the instructions executed by the processor of the computer or other programmable data processing device create means for implementing the functions / actions specified in the flowchart and / or one or more block diagram frames.

[0106] These computer program instructions may also be stored in a computer-readable memory, which may instruct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions / actions specified in the flowchart and / or block diagram.

[0107] The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions / actions specified in the flowchart and / or block diagram.

[0108] The flow chart and block diagram in the figure can illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a part of a module, segment or code, which includes one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the function indicated in the box may not occur in the order indicated in the figure. For example, two boxes displayed in succession may actually be executed substantially at the same time, or may sometimes be executed in the opposite order, not executed at all, or executed in any combination with any other flow chart, depending on the function involved. It should also be noted that each box in the block diagram and / or flow chart and the combination of boxes in the block diagram and / or flow chart can be implemented by a special hardware system or a combination of special hardware and computer instructions that performs a specified function or action.

[0109] The terms used herein are only used to describe specific embodiments and are not intended to limit the present disclosure. The singular forms "a", "an" and "the" used herein are also intended to include plural forms, unless the context clearly indicates otherwise. It should be further understood that when the terms "comprise" and / or "comprising" are used in this specification, the presence of the features, integers, steps, operations, elements and / or components is specified, but the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof is not excluded.

[0110] The corresponding structures, materials, acts, and equivalents of all means or step plus function elements in the following claims are intended to include any structure, material, or act for performing the function in combination with other claimed elements as specifically claimed. The description of the present disclosure is presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the disclosure in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the present disclosure. The embodiments are selected and described in order to best explain the principles and practical applications of the present disclosure and to enable others of ordinary skill in the art to understand the various embodiments of the present disclosure with various modifications suitable for the intended particular use.

[0111] A number of embodiments have been described. Having thus described the disclosure of this application in detail and by reference to examples thereof, it will be apparent that modifications and variations are possible without departing from the scope of the disclosure as defined in the appended claims.

Claims

1. A computer-implemented method, executed on a computing device, comprising: receiving an input voice signal; generating a transcription of the input speech signal via an automatic speech recognition (ASR) system; identifying, from the transcription, one or more split points between one or more sensitive content portions and one or more non-sensitive content portions; as well as The input speech signal is segmented into the one or more sensitive content portions and the one or more non-sensitive content portions based at least in part on the one or more segmentation points, thereby defining one or more sensitive content signals and one or more non-sensitive content signals.

2. The computer-implemented method of claim 1 , wherein the one or more sensitive content portions include one or more of the following: Personally Identifiable Information (PII); and Protected Health Information PHI.

3. The computer-implemented method of claim 1 , further comprising: One or more attributes of the input speech signal are modified to mask personally identifiable information within the input speech signal.

4. The computer-implemented method of claim 1 , further comprising: The one or more sensitive content signals are provided to a first transcriptionist.

5. The computer-implemented method of claim 4, wherein providing the one or more sensitive content signals to the first transcriptionist comprises providing each respective one of the one or more sensitive content signals to a different transcriptionist among a plurality of transcriptionists.

6. The computer-implemented method of claim 4, further comprising: The one or more non-sensitive content signals are provided to a second transcriptionist.

7. The computer-implemented method of claim 6, further comprising: receiving a manually generated transcription of the one or more sensitive content signals from the first transcriptionist, thereby defining a sensitive content transcription; receiving a manually generated transcription of the one or more non-sensitive content signals from the second transcriptionist, thereby defining a non-sensitive content transcription; as well as The sensitive content transcript is combined with the non-sensitive content transcript to define a combined transcript.

8. The computer-implemented method of claim 1 , further comprising: A speech processing model is trained based at least in part on the combined transcription and the input speech signal.

9. A computer program product residing on a non-transitory computer readable medium having a plurality of instructions stored thereon, the plurality of instructions, when executed by a processor, causing the processor to perform operations comprising: receiving an input voice signal; generating a transcription of the input speech signal via an automatic speech recognition (ASR) system; identifying, from the transcription, one or more split points between one or more sensitive content portions and one or more non-sensitive content portions; as well as The input speech signal is segmented into the one or more sensitive content portions and the one or more non-sensitive content portions based at least in part on the one or more segmentation points, thereby defining one or more sensitive content signals and one or more non-sensitive content signals.

10. The computer program product of claim 9, wherein the one or more sensitive content portions include one or more of the following: Personally Identifiable Information (PII); and Protected Health Information PHI.

11. The computer program product of claim 9, wherein the operations further comprise: One or more attributes of the input speech signal are modified to mask personally identifiable information within the input speech signal.

12. The computer program product of claim 9, wherein the operations further comprise: The one or more sensitive content signals are provided to a first transcriptionist.

13. The computer program product of claim 12, wherein providing the one or more sensitive content signals to the first transcriptionist comprises providing each respective one of the one or more sensitive content signals to a different transcriptionist among a plurality of transcriptionists.

14. The computer program product of claim 12, wherein the operations further comprise: The one or more non-sensitive content signals are provided to a second transcriptionist.

15. The computer program product of claim 14, wherein the operations further comprise: receiving a manually generated transcription of the one or more sensitive content signals from the first transcriptionist, thereby defining a sensitive content transcription; receiving a manually generated transcription of the one or more non-sensitive content signals from the second transcriptionist, thereby defining a non-sensitive content transcription; as well as The sensitive content transcript is combined with the non-sensitive content transcript to define a combined transcript.

16. The computer program product of claim 9, wherein the operations further comprise: A speech processing model is trained based at least in part on the combined transcription and the input speech signal.

17. A computing system comprising: Memory; and A processor configured to receive an input speech signal, wherein the processor is further configured to generate a transcription of the input speech signal via an automatic speech recognition (ASR) system, wherein the processor is further configured to identify one or more segmentation points between one or more sensitive content parts and one or more non-sensitive content parts from the transcription, wherein the processor is further configured to segment the input speech signal into the one or more sensitive content parts and the one or more non-sensitive content parts based at least in part on the one or more segmentation points, thereby defining one or more sensitive content signals and one or more non-sensitive content signals, wherein the processor is further configured to provide the one or more sensitive content signals to a first group of transcriptionists among a plurality of transcriptionists, wherein providing the one or more sensitive content signals to the first group of transcriptionists includes providing each corresponding sensitive content signal of the one or more sensitive content signals to a different transcriptionist among the plurality of transcriptionists, and wherein the processor is further configured to provide the one or more non-sensitive content signals to a second group of transcriptionists.

18. The computing system of claim 17, wherein the one or more sensitive content portions include one or more of the following: Personally Identifiable Information (PII); and Protected Health Information PHI.

19. The computing system of claim 17, wherein the processor is further configured to: One or more attributes of the input speech signal are modified to mask personally identifiable information within the input speech signal.

20. The computing system of claim 17, wherein the processor is further configured to: receiving manually generated transcriptions of the one or more sensitive content signals from the first group of transcribers, thereby defining sensitive content transcriptions; receiving manually generated transcriptions of the one or more non-sensitive content signals from the second group of transcribers, thereby defining non-sensitive content transcriptions; as well as The sensitive content transcript is combined with the non-sensitive content transcript, thereby defining a combined transcript.

Images