2023.03.01

AI Ethical Guidelines for Voice Experiences: A Case for Inclusivity and Trustworthiness - Version 2.0

The Open Voice Network

What are voice-specific issues of ethical and moral concern, and how do we define them?
What are the values to be promoted for ethical interaction with voice technology?
What are the rights to be respected for successful, ethical voice experiences?
What preventative measures can we take to protect people from voice-specific harm?
What can we do to cover gaps in voice that are not yet fully covered by existing laws and guidelines?
What can we learn from past voice-specific harmful incidents and vulnerabilities, and how can we prevent them from happening in the future?

This paper outlines a voice-specific ethical framework based on fundamental ethical principles of Compliance, Transparency, Inclusivity, Privacy Protection, Accountability, and Sustainability throughout the design process to avoid potential harms such as bias, discrimination, and exclusion for humans interacting with voice technology. Next steps include industry collaboration and the development of a TrustMark initiative based on these guidelines.

Updates to V2.0 of this document include:

An updated introduction and voice data analysis section
New guidance on opt-in policies and designing responses for emergency situations

For the latest, detailed information on the work of the Open Voice Network, please visit our website at openvoicenetwork.org and the Open Voice Network GitHub Repository at github.com/open-voice-network/docs.

2.0 INTRODUCTION

2.1 TARGET AUDIENCE

These ethical guidelines for voice experiences are for enterprises and organizations using or considering using voice technology. In this document, the Open Voice Network’s Ethical Use Task Force outlines a voice-specific ethical framework designed to reassure users of the full respect of their rights.

2.2 GOAL

Our goal is to increase awareness of voice technology capabilities, explore ethical benefits and challenges, and reduce fear of adverse effects.

2.3 THERE IS MUCH AT STAKE

The recent explosion of generative AI and large language models such as GPT-4 have raised awareness and accelerated the adoption of conversational AI across all verticals, especially within enterprises. They can answer queries, complete sentences, write entire articles, and more. Natural language processing (NLP) inside these models opens up new possibilities for voice integration. Although generative AI and large language models offer many benefits, they create new ethical risks and potential harms. They can simulate human intelligence and provide contextual answers to typed queries, but they lack human understanding and sometimes provide false facts. With these new tools being used in everyday life, ethical behavior and implementation of ethical standards and guidelines are essential.

2.4 WHAT'S NEXT?

Our next steps include promoting these guidelines and developing the Open Voice Network TrustMark initiative. The guidelines will be promoted through webinars, conference talks, outreach, and other events. The TrustMark initiative will help enterprises and organizations turn these ethical guidelines into actionable steps to take to create ethical voice experiences. We will also update this document on a regular basis to address ethical ramifications of voice technology advancements.

2.5 ABOUT THE ETHICAL USE TASK FORCE AND OUR APPROACH

The Ethical Use Task Force is a group within the Open Voice Network dedicated to the ethical use of voice technology. We focus on voice as a subset of conversational AI, which is defined by The Open Voice Network (2022b, p. 57) as “the set of technologies to enable automated communication between computers and humans." Our members approach voice technology from every angle, from designers and practitioners to audio experts, community organizers, voice actors, academics, entrepreneurs, and users. Over the course of 2.5 years, the Ethical Use Task Force met weekly to review, discuss, analyze, and interpret reputable sources of information related to voice and ethics, attended ethics-related events, and brought in experts from various fields to share their knowledge and resources about ethics. This document reflects our discussions, research, and analysis of information brought up in our meetings, the events we attended, and our individual research. Our ethical principles reflect a belief in a partnership between humans and technology, where technology enhances human capabilities rather than replaces them and contributes to the well-being of individuals, society, and the environment. The same ethical principles apply to the use of synthetic voice in voice technology. Additional guidance can be found in the Open Voice Network report, “Synthetic Voice For Content Owners and Creators.”

3.0 WHO OWNS VOICE DATA?

“The creator of voice data is the owner of that data unless there is explicit consent to the transfer of ownership” (Open Voice Network, 2023).

4.0 THE OPEN VOICE NETWORK WORKING DEFINITION OF PERSONAL VOICE DATA

Direct or inferred information that can identify an individual through the sound of their voice such as a voiceprint, emotional state, and health status
Personal information spoken out loud by an individual
Ambient noise
Personal information linked to an individual’s voice such as account numbers, names, contact information, contacts, and device information

5.0 VOICE DATA ANALYSIS

5.1 HOW IT WORKS

The human voice is composed of distinctive sonic characteristics unique to each individual. The analysis of the voice focuses on the sound used to produce it, defined by Abercrombie (1967) as “those characteristics which are present more or less all the time that a person is talking” (as cited in Laver, 1980, p. 1). These include tone, speed, duration, volume, and pitch.

5.2 BENEFITS AND USE CASES

Voice technology can identify, analyze, and deduce sensitive and personal information about an individual from these distinctive characteristics in the sound of their voice and ambient noise. This can include information related to an individual’s identity such as their name, location, socioeconomic status, physical characteristics, physical health, mental health, emotions, intent, and more.

Figure 2
***** Data that can be processed and analyzed from the sound of a voice *****

Note. A graphic describing how voice technology can identify, analyze, and deduce sensitive personal information related to an individual’s identity, physical characteristics, physical health, mental health, socioeconomic status, location, emotions, intent, and more. By Open Voice Network, 2022.

These characteristics can be processed and analyzed for various purposes including:

Authentication
Evaluation
Disease detection
Assessing emotional state (Chen, 2019)

5.2.1 AUTHENTICATION

Voice biometric authentication verifies an individual’s identity through their unique voice characteristics and speech patterns (Jiřík, 2021). A digital recreation of the voice characteristics and patterns, called a voiceprint, is created and stored in a database (Sivakumar, 2020). Whenever you need to be authenticated, your voice is matched to the stored voiceprint to confirm your identity (Sivakumar, 2020).

Voice authentication can be used for purposes such as fraud detection, call center identification verification, contactless login, two-factor authentication, financial transactions, digital signatures, workforce management, and more (Pinto, 2021). Benefits include saving time and money, increasing security and accessibility, and ease of implementation (Sivakumar, 2020).

While voice authentication adds extra layers of security protection, it is not foolproof. Bias and discrimination are among potential issues. Accuracy can vary depending on ambient noise level and changes to the voice such as illness, accents, and speech patterns (Pinto, 2021). According to Pinto (2021), voice data can be replicated if servers and databases are hacked. Fraud can also occur through the use of synthetic voice (Open Voice Network, 2022a, p. 12). For more on voice-specific security, reference our Data Security Specific to Voice whitepaper (Open Voice Network, 2022a).

5.2.2 EVALUATION

Voice technology can be a factor in determining whether people are qualified for jobs, loans, housing, and more. While automating these processes can save enterprises and organizations time and money, it can lead to bias and discrimination in the decision-making process. Voice analysis can generate a complete personality profile based only on the sound of the voice, which can then correlate the profile to sensitive categories such as race, ethnicity, or gender. For example, as more companies use voice technology to help facilitate hiring/interview decisions, a candidate could be denied a job due to algorithms favoring specific attributes. According to Burns (2021), “...[S]oftware like AON’s vidAssess scans a recorded video interview for spoken words and phrases that align with the competencies you are looking for in an employee and scores them for an at-a-glance overview,” although “early signals suggest certain accents and tones are being prioritized” over others (paras. 16 & 20). In a study conducted by Dr. John Baugh (2002), landlords were more likely to deny housing opportunities to prospective tenants of color based solely on the sound of their voice, regardless of their qualifications. Baugh found that this type of “linguistic profiling” occurred when searching for employment, service providers, and loans as well (Rice, 2002, para. 2).

5.2.3 DISEASE DETECTION

In the medical field, specific features and signals in the sound of the voice known as biomarkers can potentially be analyzed to detect, treat, monitor, and prevent diseases such as:

Coronary artery disease.
Neurological disorders, such as Amyotrophic lateral sclerosis (ALS), Parkinson’s, and
Alzheimer’s.
Psychiatric disorders, such as schizophrenia, bipolar disorder, post-traumatic stress disorder, and depression (Fagherazzi et al., 2021).

Current research uses voice biomarkers to detect Covid-19 in individuals through the analysis of their voice and respiratory sounds, such as coughing and breathing (Despotovic et al., 2021). One Covid-19 study looking at cough patterns achieved 88.52% accuracy, 88.75% sensitivity, and 90.87% specificity (Despotovic et al., 2021).

Voice analysis also shows encouraging results in detecting Parkinson’s disease. In a study conducted to determine if algorithms could accurately detect Parkinson’s disease in individuals, 6,500 sound samples were collected from a group of around 6,000 people (Wroge et al., 2018). Algorithms were then used to determine which of those people had Parkinson’s and which people did not, with a peak accuracy rate of 85% (Wroge et al., 2018).

However, as with other biomarkers, voice biomarker analysis currently comes with concerns of accuracy, reliability, specificity, sensitivity, variability, bias, privacy, and interpretation, especially when it comes to something as serious as diagnosing diseases (Mayeux, 2004). To date, no digital voice biomarker technology has been approved by the United States Food and Drug Administration (FDA), so it is important not to rely solely on voice analysis when making important medical decisions (Wroge et al., 2018).

5.2.4 ASSESSING EMOTIONAL STATE

In voice experiences, both spoken words and the sound of the voice can be analyzed to determine an individual’s emotional state. Sentiment analysis determines how people are feeling based on the words they use, and speech emotion recognition (SER) determines how people are feeling based on the sound of their voice (Ingale & Chaudhari, 2012). Emotional states such as happiness, sadness, fear, and anger can potentially be detected as well as physiological states, such as intoxication and hunger (Maghilnan & Kumar, 2017).

Speech emotion recognition and sentiment analysis can be used for personalizing experiences, marketing, learning, clinical studies, remote patient monitoring, surveillance, entertainment, banking, and much more (Kerkeni et al., 2020). However, extra caution should be used when relying solely on speech emotion recognition and sentiment analysis.

The information that can be inferred from these emotions is highly sensitive and personal. Making decisions based on this subjective information can cause situations to escalate. Emotions are tricky to interpret, as they are often temporary and can easily be misconstrued for many reasons, including cultural differences.

5.2.5 RISKS

Voice characteristics as identifiers raise several concerns regarding consent, data usage, data ownership, data security, and data accuracy. Collecting and using biometric data is already regulated in many countries, international unions such as the European Union, and several US states. However, current anti-discrimination laws do not fully cover discrimination from the sound of the voice.

Many ethical issues surrounding voice analysis run into broader questions, such as:

When is it okay to use the information?
How much information is acceptable to use?
Is the information accurate?
What constitutes discrimination or bias?

While not unique to voice alone, misuse of personal data, poor data security management, and training with limited datasets can result in personal data leakage, personal data theft, and unfair outcomes. Poor quality training data, limited in size and scope, can lead to frustrating experiences, exclusion, bias, and systemic discrimination.

If not properly constrained, this information can be used to make decisions and recommendations regarding people’s lives that could potentially cause irreparable harm. Prison convictions have been made based on voice analysis, only to be overturned years later because of contradictory DNA evidence (Claypoole, 2021).

Figure 1

Conversational/Voice AI Risks

Note A figure identifying the risks of conversational/voice AI including privacy, control, security, economic, performance, and societal risks. By Open Voice Network, 2022.

These types of decisions and outcomes can be prevented and mitigated if the proper measures and guidelines are in place.

6.0 ETHICAL FRAMEWORK FOR TRUSTWORTHY VOICE EXPERIENCES

Figure 3 Open Voice Network ethical principles for voice experiences

Note. Overarching principle of governance and compliance encompassing the other ethical principles of transparency, privacy protection, inclusivity, accountability, and sustainability to achieve trustworthy voice experiences. By Open Voice Network, 2022.

6.1 GOVERNANCE/COMPLIANCE

Overarching compliance with all of the ethical principles in this document, standards, guidelines, and existing laws and regulations will be crucial to achieving trust in voice experiences. We accept that achieving an expected outcome is crucial to gaining trust. However, it is equally— if not more important— to do the following:

Uphold and comply with the highest standards and the law
Maintain open and transparent communication
Minimize the amount of personal voice data collected
Continuously monitor and audit decisions and outcomes regarding personal voice data

This will help reduce any uncertainty and vulnerability people might experience before, during, and after any voice interaction.

6.1.1 FOLLOW EXISTING LAWS AND REGULATIONS

Thoroughly inform yourself of existing laws and regulations wherever people will potentially use your voice experience, as there are different laws and regulations for voice experiences depending on your location.

Existing laws and regulations include but are not limited to:

The California Consumer Privacy Act (CCPA) of 2018,
European General Data Protection Regulation (GDPR),
European Accessibility Act,
Section 508 of the Rehabilitation Act of 1973,
Illinois Biometric Information Privacy Act (BIPA),
Health Insurance Portability and Accountability Act (HIPAA), and
Children’s Online Privacy Protection Rule (COPPA).

To track US state privacy legislation, reference the Husch Blackwell 2023 State Privacy Law Tracker: https://www.huschblackwell.com/2023-state-privacy-law-tracker

6.1.2 USE INDUSTRY STANDARDS, GUIDELINES, AND BEST PRACTICES WHEN DEVELOPING VOICE TECHNOLOGIES AND EXPERIENCES

As voice-specific laws and regulations are still evolving, it is important to implement industry standards, guidelines, and best practices such as the ones in the European Data Protection Board guidelines on virtual voice assistants. This will increase your chances of being compliant with existing laws and regulations and prevent potential compliance issues from occurring in the future.

6.1.3 CONDUCT REGULAR INTERNAL AND EXTERNAL AUDITS

Audits specifically related to personal voice data should be conducted to monitor ethical decisions and outcomes, measure success if key performance indicators have been set, and ensure that laws, standards, and guidelines are being followed.

6.1.4 HIRE OR CONSULT ETHICISTS THAT CAN REVIEW AND EVALUATE THE IMPACT OF ETHICAL DECISIONS BEING MADE REGARDING PERSONAL VOICE DATA

Working with experts that can review and evaluate ethical decisions can help prevent potential issues and “enable the creation of well-defined policies that promote human rights, safety, and economic benefits” (IEEE, 2019, p. 206).

6.2 TRANSPARENCY

Open and transparent communication with users is key to building a foundation of trust. Any decisions regarding personal voice data should be disclosed and presented in an easily accessible, understandable, and explainable format.

Should the intent and purpose of collecting personal voice data become vague or nonexistent at any point, trust will erode rapidly. Lack of transparency in the past has led to backlash against major tech companies for sending audio recordings from voice assistants to contractors for review without anyone’s knowledge (Barrett, 2019). People are vulnerable every time they interact with voice assistance, so transparency will encourage them to trust that no harm will come from using voice technology.

6.2.1 BE CLEAR ABOUT VOICE DATA OWNERSHIP

Human beings own their voice data unless they explicitly consent to the transfer of ownership. Transfer of ownership should be clearly and explicitly stated in the terms and conditions, and all entities assuming ownership should be listed.

There cannot be any ambiguity when it comes to the ownership of voice data for the protection of everyone involved. If ownership is transferred, people should clearly understand who owns what voice data and why.

These guidelines address voice data ownership only for living human beings. Future research will be done to address the impact of voice data ownership for the deceased.

6.2.2 INTEGRATE AN 'ETHICAL BLACK BOX' IN THE DESIGN

UNI Global Union (2017) suggests one way to enhance transparency is to integrate and implement an “ethical black box,” similar to one found in airplanes (p. 7). This black box should be able to record all decisions made and explain the reasoning behind all decisions. In robots, this applies to movement and sensory data as well. The read-out of the decisions and reasoning behind them should be easily understandable, concise, and simple.

Amazon has already started doing this with Alexa. People can ask Alexa, “Why did you do that?” and Alexa will explain why a specific action was taken (Gershgorn, 2019). This technology is still evolving, but understanding what happened and why certain decisions were made will improve the overall experience and strengthen the relationship of trust in technology.

6.2.3 DISCLOSE THE TYPE OF VOICE ANALYSIS USED AND THE PURPOSE OF USING IT

Make sure people are informed of the type of voice analysis being used, the purpose of using that type of voice analysis, and how the voice data collection and analysis work.

Examples of voice analysis include:

Voice biometrics
Speech emotion recognition
Voice biomarkers

6.2.4 DISCLOSE THE TYPE OF PERSONAL VOICE DATA BEING COLLECTED AND HOW IT IS BEING USED

During voice interactions, people reveal more personal information about themselves through their voice than they might realize, even for seemingly low-risk interactions such as asking for the weather or playing music. From only one interaction, no matter how small, platform vendors can choose to collect data that includes:

Personal information, including your name, address, and location of your device.
Personal interests and personal description stored on your profile.
Images and videos stored on your account.
An acoustic model of voice characteristics.
Phone numbers, addresses, and emails of your contacts.
Information about your devices, such as specs and configuration.
Devices linked to your smart assistant.
A record of interactions and requests made through your smart assistant.
Internet provider and connectivity information, and more (Cohen, 2021).

While we acknowledge that collecting data is an integral part of improving a voice experience, people should know what type of personal voice data is being collected, why it is being collected, how it is being managed, and by whom. For example, if voice biomarkers are collected for medical purposes, explain:

what voice biomarkers are,
who is collecting them,
why they are being collected,
how they are collected and stored,
how long they will be stored,
which specific ones are being analyzed and for what purpose, and
how the analysis will be interpreted.

People should also be informed when both speech emotion recognition and sentiment analysis are run, as they both affect the impact of the data and what people are willing to share.

6.2.5 DISCLOSE THE PURPOSE FOR EVERY PIECE OF PERSONAL VOICE DATA COLLECTED

Every piece of personal voice data needs to be accounted for and have a clear purpose that is disclosed in the terms and conditions provided to the user.

People need to understand the benefits they get out of giving away sensitive voice data and why it is being used. These benefits should outweigh the harms, given how much can be collected and inferred from personal voice data.

6.2.6 DISCLOSE ALL THIRD PARTIES INVOLVED WITH ACCESSING, OWNING, AND ANALYZING PERSONAL VOICE DATA

List the names of the third parties, why they need access to the data, what they will use the data for, and how the data will be processed and analyzed. Provide links to each party’s privacy policy and terms and conditions.
If a new party needs access to the personal voice data after consent is given, consent needs to be provided again.

6.2.7 CLARIFY WHEN DEVICES ARE LISTENING VS. RECORDING

A major deterrent for the use of voice technology is the concern that devices are spying or listening in on private conversations at all times. Disclose when devices are listening, what they are listening for, what triggers a voice device to start recording, and exactly when recordings are started and stopped.

6.2.8 HAVE A PROCESS IN PLACE FOR THE WITHDRAWAL OF CONSENT FOR THE USE OF PERSONAL VOICE DATA

Regardless of an opt-in or opt-out consent policy, there should be an explicit and well-defined procedure to withdraw consent for the use of personal voice data at any time, in accordance with existing laws and regulations. There should also be a clear explanation of what happens to personal voice data after consent is withdrawn.
Withdrawing consent should be an easy and straightforward process with a maximum of two pieces of information:

The withdrawal request (mandatory)
The reason for withdrawing consent (optional)

6.2.9 DISCLOSE HOW AND WHERE PEOPLE CAN DELETE THEIR PERSONAL VOICE DATA

The process for deleting personal voice data should be featured prominently and stated clearly in the terms and conditions, in accordance with local, state, federal, and international laws. The information should include how and where the voice data is deleted, how people can confirm that their voice data was deleted, and how the deletion is handled when third parties are involved. This is particularly important for accidental data processing.

6.2.10 WRITE TERMS AND CONDITIONS AND PRIVACY POLICIES ABOUT PERSONAL VOICE DATA CLEARLY SO PEOPLE OF ALL LITERACY LEVELS CAN UNDERSTAND THEM

Given the sensitive personal nature and complexity of the human voice and voice experiences, it is essential that all people clearly and easily understand what they are consenting to before they interact with voice technology.

6.3 PRIVACY

The Universal Declaration of Human Rights by the United Nations (1948) and many nations worldwide recognize privacy as a fundamental human right. Ensuring the privacy and protection of privacy of all people should be a priority from conception to the testing and launch phases, no matter where you are located.

Figure 4

Open Voice Network privacy principles

Note. An illustration and explanation of the Open Voice Network privacy principles including transparency, consent, limited collection and use, and control. By Open Voice Network, 2022.

Privacy protection for voice experiences is crucial for the following reasons:

The biometric nature of voice and the amount of sensitive information that is shared with every voice interaction
Risk of accidental data processing
Voice data that may contain sensitive information must be protected under existing laws and regulations
Security vulnerabilities with voice technology could lead to hackers or other third parties controlling voice-enabled devices, listening to and recording private audio conversations at any time, extracting personal voice data, and leaking or selling personal information (American Bar Association, 2016). Any breaches in security could negatively impact trust in voice technology and cause serious liability issues.

6.3.1 MINIMIZE THE AMOUNT OF PERSONAL VOICE DATA BEING COLLECTED

For privacy and security purposes, do not collect personal voice data unless there is a specific reason for collecting it. If it is essential to collect personal voice data, gather only the minimum amount of voice data needed to fulfill a specific purpose, nothing more. Get to the root of why a particular piece of personal voice data is being collected, then figure out if it is worth collecting for that purpose. If more personal voice data is needed in the future, you can ask for consent again when that data is needed.

6.3.2 PRIORITIZE DATA SECURITY

If personal voice data needs to be stored for a specific reason, it should be stored safely and securely. If any sensitive voice data is processed, managed, or stored by third parties, links to all third-party privacy policies should be provided. If a data breach occurs, users should be notified immediately. Detailed investigations and records should be kept regarding any breaches experienced by an organization in case users want information and explanations regarding the risks to which they have been exposed.

6.3.3 USE OPT-IN POLICIES INSTEAD OF OPT OUT POLICIES

An opt-in policy requires consent before any personal data can be collected, as opposed to an opt-out policy, which collects personal data by default unless users specifically choose to opt out of data collection (Securiti 2019). While an opt-in approach might limit widespread and diverse data collection, it is the more ethical approach to take. According to The Editorial Board (2021), because most companies use opt-out policies, “... consumers’ data is now so widely dispersed that security breaches can sweep up information about people who’ve never even visited certain websites.”

Collecting the minimal amount of voice data necessary and being fully transparent about what’s happening with personal voice data being collected, processed, and analyzed will go a long way towards getting as many people to consent as possible with an opt-in policy.

6.3.4 THOROUGHLY VET THIRD PARTIES THAT ARE PROCESSING AND HANDLING PERSONAL VOICE DATA

Pay particular attention to people’s privacy rights when engaging in third-party negotiations, especially when children could be involved. Make sure that the highest standards of privacy for voice experiences, especially the ones processing personal voice data, are guaranteed by both organizations.

Major tech companies typically send audio files to a mix of internal employees and contractors for review (Day et al., 2019). In the past, these companies have received backlash for not explicitly letting users know that humans were reviewing their conversations and ended up issuing rare apologies and making changes to their privacy settings and policies because of it (Nieva & Rubin, 2019). Although the purpose of human review is to train the algorithms and improve the experience, audio snippets can be tied back to individuals and privacy can be compromised if the audio is not properly anonymized.

Recordings from one tech company sent to contractors were tied to an account number, the user’s first name, and the serial number of the device (Day et al., 2019). Contractors were able to hear highly personal information such as “lonely sounding people confessing intimate secrets and fears; a boy expressing a desire to rape; men hitting on Alexa like a crude version of Joaquin Phoenix in Her...Other contractors recall hearing kids share their home address and phone number, a man trying to order sex toys, a dinner party guest wondering aloud whether Amazon was snooping on them at that very instant” (Carr et al., 2019, para. 3).

We understand that human review of conversations is vital to improving voice experiences, but it cannot come at the expense of people’s privacy. The harms that can occur from the leakage of personal voice data or the mishandling of information from third parties or unknown parties could be devastating. One company had to briefly stop all human review of audio from voice interactions after it was found that a contractor had been leaking parts of the audio (Newman, 2019).

Best practices for organizations and enterprises include:

Giving people an option to opt in to potential human review
Ensuring all audio snippets cannot be linked back to the user
Setting the expectation that all third-party contractors will adhere to the same ethical standards that enterprises and organizations expect of themselves and their employees

6.3.5 DELETE PERSONAL VOICE DATA WHEN IT HAS FULFILLED ITS PURPOSE

Do not store personal voice data longer than necessary. The longer it is stored, the greater the risk of something negative occurring. Fully delete personal voice data in accordance with existing laws after it has fulfilled its purpose, which should be stated clearly in terms and conditions.

6.3.6 ALLOW PEOPLE TO PERSONALLY ACCESS AND DELETE THEIR VOICE DATA AT ANY TIME

People should be able to request the deletion of part or all of their personal voice data at any time in accordance with local, state, federal, and international laws. This process should be timely, easy, and straightforward to understand. People should also receive confirmation that their data was deleted successfully.

6.3.7 HAVE A PROCESS IN PLACE FOR ACCIDENTAL DATA PROCESSING

Personal voice data can be recorded, stored, and processed without knowledge and consent if a device is accidentally activated by a word similar to its wake word (Schönherr et al., 2020). That can cause serious privacy issues as recordings are sent to the cloud and can potentially be processed and analyzed.

Lawsuits have been filed against Google, Amazon, and Apple, claiming that devices were activated and recorded private conversations even though the wake word wasn’t used (Lerman, 2021). According to Lerman (2021), those recordings were then allegedly used to target advertisements and given to third parties, which is a clear violation of privacy. We recognize that accidental voice data processing is not always easy to catch, which is why a clear process for users and providers to continuously monitor, audit, and delete accidental voice data processing is important.

6.4 INCLUSIVITY

Interactive technologies are becoming increasingly voice-enabled, creating more opportunities to make products and services accessible for everyone and give people access to the digital world. In particular, marginalized or underserved groups can integrate voice technology into their daily activities without depending on others for help. Because of its inclusive nature, voice can have a real impact and bring tangible improvements to people’s lives.

Many people rely on voice technology to complete daily tasks such as keeping in touch with others, interacting with other Internet-enabled devices, and searching for information (Ammari et al., 2019). Others rely on voice technology for accessibility reasons. Take Kaden, for example.

Kaden is a boy with cerebral palsy who cannot walk or talk, is legally blind, and can use only the pinky finger on his left hand (Snow, 2019). Kaden uses a speaking device to communicate with his Amazon Echo Show by pressing buttons with his pinky to verbalize preselected words or sentences. His Echo Show was set up so that Kaden could ask for things such as the news or a joke and keep in touch with family, but Kaden has been using the technology beyond its intended use, calling his grandfather one evening through his Echo Show to ask if they could go for a car ride.

Voice assistance is also used for critical tasks such as calling for help in an emergency or connecting to a hotline if someone is in dire physical and mental distress (Graham, 2017). It is particularly useful in situations when hands or eyes are busy, such as cooking or driving. Language is becoming less of a barrier, as audio can potentially be translated from text to speech in any language.

Initiatives such as Translators Without Borders are leveraging speech technology to address the needs of refugees and fight disinformation regarding COVID-19 in Nigeria and Congo (TWB Communications, 2020). One of their long-term programs aims at creating a more balanced communication with populations in low-literacy contexts that have historically been marginalized.

By engaging with and learning from a diverse group of people, we can work towards making products and services more accessible through voice and including everyone in a voice-enabled future. Voice technology systems should also be assessed regularly to ensure they are fair, free of bias, and designed with the input and participation of a diverse range of users.

6.4.1 PROVIDE MULTIPLE OPTIONS FOR CONSENTING TO TERMS AND CONDITIONS AND PRIVACY POLICIES IN ACCESSIBLE FORMATS

Consent should be available in multiple modalities and languages for accessibility and ease. Voice alone will exclude some categories of people from expressing consent. Anyone should be able to give legal consent to use personal voice data and agree to the terms and conditions and privacy policies in accordance with local, state, federal, and international laws.

6.4.2 ASSEMBLE DIVERSE TEAMS TO DESIGN AND DEVELOP VOICE EXPERIENCES

Diversity within teams is the first step toward preventing societal harms, such as favoring one group arbitrarily over another (Li, 2020). Well-rounded teams can also minimize potential bias because of the range of experiences and perspectives that can be drawn from throughout the entire design process (Orduña, 2019). This is important, especially for speech-recognition tools.

There is already known bias against women in general and people with accents that use voice technology in cars (McMillan, 2011), and racial bias against Black individuals, as proven by a study that tested commercial speech-to-text tools used by five of the major tech companies (Koenecke et al., 2020). Diverse teams will be able to spot and anticipate these types of scenarios and think more creatively, which will lead to more inclusivity in voice experiences and better scalability across enterprises and organizations (Shastri, 2020).

Proper training and education of these teams is crucial as well. The more knowledge the team has about how the technology works and the potential issues that can occur, the more they can mitigate those issues and educate others about them.

6.4.3 TEST VOICE EXPERIENCES WITH DIVERSE GROUPS OF PEOPLE AT EVERY STEP OF THE DESIGN PROCESS

Feedback from diverse groups can spark new ideas, increase accessibility, uncover design issues, expose potential bias and discrimination, and create broader customer bases. It is particularly important to test wake words and invocation words. If people can’t invoke your voice assistant or skill, they will not be able to use it. Certain people might have a harder time saying specific letters or words, so make sure the words or names you use are as universally easy for people to say and remember as possible.

Also reach out to organizations, groups, and individuals that could potentially be excluded from the design due to challenges, disabilities, and underrepresentation, and include them in your testing groups. Google is currently testing Project Relate, an app that helps people with diverse speech patterns communicate with Google Assistant (Schwartz, 2021). This project will help improve Google’s speech recognition technology for people with ALS, Cerebral Palsy, Parkinson’s disease, and more. The more variation you have, the more people will be able to interact with the design.

6.4.4 GATHER LARGE, DIVERSE AND INCLUSIVE DATASETS FOR TRAINING VOICE AI MODELS

Voice AI learns from data produced by humans, incorporating human bias into voice datasets and leading to algorithms that reflect human prejudices. Those biases need to be taken into account and mitigated when collecting and processing data.

The Intelligence Community (2020) says unintentional bias can “undermine analytic validity, harm individuals, or impact civil liberties such as freedom from undue government intrusion on speech, religion, travel, or privacy” and may be “introduced through the process of data collection, feature extraction, curating/labeling data, model selection and development, and even in user training” (sec. 5, para. 2).

Awareness of these biases can go a long way toward designing more inclusive and accessible voice products and services, but that alone is not enough. In addition to education, awareness, and documenting previously known biases, determine at which points humans should be involved in the decision-making process and the level of accountability required for each role through risk assessment (Intelligence Community, 2020). Source broad and diverse voice datasets and consider obtaining feedback from experienced ethicists to guide inclusive design.

6.4.5 ENGAGE WITH AND EDUCATE YOUR COMMUNITIES

There is still a learning curve when it comes to voice interactions. It is easy for people to give up if a voice experience does not work for them. The more people know how to use voice technology and see the value, the more people can trust that it will work for them and make informed decisions about how to use it.

Engaging with your communities allows you to tailor features to human needs, learn what people want out of voice experiences, understand the pain points and how to overcome them, and figure out what might be missing. Listening to and implementing feedback will make the experience more accessible and beneficial to all. If one person says that your voice experience is not accessible, it is likely that others have the same problem. Provide a way to gather feedback and create an active feedback loop that is continuously monitored to help incorporate suggestions.

6.4.6 BE AWARE OF EXISTING STEREOTYPES WHEN APPLYING GENDER TO VOICE ASSISTANTS

Gender bias is particularly prevalent in voice assistants, often as a result of the makeup and biases of the team members creating them. In the tech industry, these teams tend to be predominantly male. Because a majority of personal assistant bots such as Siri, Alexa, and Cortana are perceived to be female, “it sends the signal that women are obliging, docile and easy-to-please helpers, available at the touch of a button or with a blunt voice command like “Hey” or “OK” (UNESCO, 2019, p. 104).

This can negatively influence the way children who interact with voice assistants on a regular basis grow up to perceive and treat women. It also opens the door for what English feminist writer, “Laura Bates (2014) calls ‘everyday sexism’: those small and seemingly harmless insinuations that women experience on a near-daily basis ... It is the kind of ‘harmless’ language that can contribute to women becoming the target of physical violence...” (as cited in Kennedy & Strengers, 2020, p. 61). This kind of abuse and harassment should be addressed through the voice assistant’s responses and used as an opportunity to raise awareness and educate people.

Other solutions for mitigating gender bias include:

Offering a variety of voices and accents instead of a singular default voice whenever possible (Robison, 2020)
Using color as a way to represent gender (Robison, 2020)
Using gender-neutral or non-binary voices like Sam, a non-binary voice created by Accenture and CereProc (Behr, 2020)

6.4.7 DESIGN APPROPRIATE RESPONSES FOR PEOPLE IN SENSITIVE OR EMERGENCY SITUATIONS

All voice experiences should be able to handle emergency or sensitive situations, no matter what type of voice experience it is. If direct help cannot be given, information for services or resources that can help should be provided. People could potentially ask for help in situations such as:

Accidents that cause physical harm and injury
Any form of abuse including physical, emotional, verbal, sexual, stalking, financial, and others (National Domestic Violence Hotline, n.d.)
Suicidal thoughts, self-harm, depression, anxiety, and loneliness
Reporting suspicious behavior (Murali, 2022)

A good approach to handling these responses, according to Murali (2022), is to acknowledge the situation, and then either direct people to an agent or provide information for services or resources such as 911 that can help them with their specific situation.

It is important that the intention of the user is clear before providing these responses, the responses are not activated accidentally, and the responses are presented in the right context in accordance with existing laws, policies, and regulations. It is also crucial that extra care is taken to protect any personal and sensitive voice data around these queries. Responses that do not adequately address these issues can lead to legal issues, lower customer satisfaction and retention rates, and less trust and awareness of the product and brand. It may not seem like these responses are your responsibility, relevant to your product or service, or worth taking time and resources to design for. However, taking the time to design these responses shows that your enterprise or organization cares about your customers and speaks volumes to your core values (Murali 2022). It will make a huge impact on those that need the help and ultimately benefit everyone in the long run.

6.4.8 AVOID USING HUMAN NAMES AS DEFAULT WAKE WORDS

The way wake words are used as commands in voice technology can impact the way humans are treated in everyday life. People named Alexa report being subject to severe bullying as a direct result of the way people interact with and treat Amazon Alexa (Ard, 2021). Some have legally changed their names because of the harassment, but for one Alexa, it got to the point where she told her mother she wanted to kill herself (Ard, 2021). The best way to mitigate this unintentional consequence is to provide a way to set custom wake words on all devices.

6.5 ACCOUNTABILITY

Accountability means maintaining and following the highest possible ethical standards when managing personal voice data and holding people accountable for their actions when negative outcomes occur.

6.5.1 BE CLEAR ABOUT THE RESPONSIBILITIES OF EACH STAKEHOLDER INVOLVED WITH THE MANAGEMENT OF PERSONAL VOICE DATA FROM THE START

Clearly defining the roles and responsibilities of each stakeholder will make it easier to hold people accountable if any negative outcomes occur as a direct result of the misuse of personal voice data.

6.5.2 PROVIDE A WAY FOR PEOPLE TO GIVE FEEDBACK IF AN UNEXPECTED OUTCOME FROM THE USE OF PERSONAL VOICE DATA OCCURS

Awareness is key to uncovering issues that might not get caught otherwise and preventing future issues from occurring.

Ways to achieve this include:

providing a form that people can fill out either online or by mail,
establishing a contact email or phone number, or
adding a flow to capture and address feedback in an existing bot.

6.5.3 ALLOW PEOPLE TO CHALLENGE OUTCOMES THAT STEM FROM THE USE OF PERSONAL VOICE DATA

If someone disagrees with an outcome that occurred because of the use of personal voice data, they should be able to challenge the outcome and get a ruling within a reasonable amount of time. If it is determined that the outcome does not need to be rectified, at least people can understand how and why the outcome happened. If it is determined that the outcome should have been different, future outcomes of the same nature can be prevented.

6.5.4 HOLD EVERY PARTY INVOLVED WITH VOICE EXPERIENCES TO THE HIGHEST LEVELS OF INTEGRITY AND ETHICAL THINKING STANDARDS

If a third party mishandles any personal voice data or does not comply with the standards and agreements set for compliance with rules and regulations, try to rectify the situation first. If the third party still does not comply, cut ties and replace it with another party that will comply with the standards.

6.5.5 TRAIN ALL TEAM MEMBERS EQUALLY, THOROUGHLY AND PROPERLY

Thorough and proper training for every employee on voice technology is essential, no matter their background or which part of the design process they are working on (Byrum, 2020). Lack of education cannot be an excuse for negative outcomes, especially regarding sensitive voice data. All team members need to “speak the same language” and “share a frame of reference” to successfully achieve their goals (Byrum, 2020, para. 10).

6.6 SUSTAINABILITY

When discussing sustainability in voice technology, we are talking about doing no harm, generating net benefits, and protecting the environment. Voice technology can “leverage quality and better standards of life and protect people’s dignity, while maintaining cultural diversity and protecting the environment” (IEEE, 2019, p. 142). Ultimately, it is about a conscious consumption of technology and voice, or engaging with technology while being aware of the larger impacts on society (Wong, 2019).

6.6.1 GENERATE NET BENEFITS FOR ALL STAKEHOLDERS

To provide net benefits for businesses, consumers, and the public sector, voice technology must have “universal respect for human rights and human dignity, the rule of law, justice, equality and nondiscrimination; ... respect for race, ethnicity, and cultural diversity; and ... equal opportunity permitting the full realization of human potential and contributing to shared prosperity” (United Nations, 2015, para. 8).

6.6.2 BE AWARE OF THE POTENTIAL HARMS VOICE TECHNOLOGY COULD INFLICT ON THE ENVIRONMENT

“Put simply: each small moment of convenience—be it answering a question, turning on a light, or playing a song—requires a vast planetary network, fueled by the extraction of non-renewable materials, labor, and data” (Crawford & Joler, 2018, para. 8).

Voice AI training models “come with an economic trade-off, as they depend on the availability of exceptionally large computational resources that necessitate substantial energy consumption” (Bheemaiah & Esposito, 2021, para. 3). Even something as seemingly simple and innocent as “What’s the weather?” requires a robust network of resources to answer.

Do thorough research and be aware of the overall cost for designing and developing voice experiences. All stakeholders should be educated on the implications of using this technology. The people using voice technology have the power to shape how it is being used, but they have to have sufficient knowledge of how the technology works to make informed decisions about how to use it.

6.6.3 CONTINUOUSLY SET GOALS AND MEASURE SUCCESS

Create key performance indicators for sustainability that are continuously monitored and measured to see if the technology has the expected impact. An example of a goal could be driving down the energy consumption of call centers by increasing their efficiency and reducing the number of people who rely on them.
Measuring the success of these goals should indicate how to move toward a more sustainable future through voice technology.

6.6.4 MAKE DECISIONS ON THE TYPE OF VOICE TECHNOLOGY USED BASED ON SUSTAINABILITY

Choose technology, implementation, and deployment based on your sustainability goals and values. For example, Haque et al. (2020) talk about an eco-friendly solar-powered and voice-controlled wheelchair built to help people with physical disabilities improve their self-mobility.
It might take a little innovation and creativity to find solutions that do not harm the environment, but there are options and resources out there.

6.6.5 ALIGN WITH ORGANIZATIONS THAT HAVE THE SAME SUSTAINABILITY VALUES AND GOALS

When collaborating with other organizations and enterprises, make sure they have similar sustainability goals. Partner with carbon-neutral companies when developing voice experiences or organizing awareness campaigns to educate people about the importance of sustainability.

6.6.6 LEAD BY EXAMPLE

Show people what is possible when sustainability is included in the voice technology decision-making process. If you can model it successfully, others will follow.

7.0 CALL TO ACTION

7.1 START SMALL

Implement small changes wherever you can at first, and then work your way up. It might be hard to see results right away, but results will show over time if you consistently implement changes.

7.2 RAISE AWARENESS ABOUT VOICE TECHNOLOGY AND CONTINUOUSLY EDUCATE YOURSELF AND OTHERS

Voice technology is continuously growing and evolving. Keeping up with the latest news and updates, raising awareness about what voice technology can do for people, and educating people on the benefits and harms of voice technology are critical for its development and growth. Over time, put good processes and practices in place to ensure voice technology is developed in a positive and ethical way. This will allow people to adopt it and use it to its fullest.

7.3 CONTINUOUSLY MONITOR AND ASSESS THE STATE OF VOICE TECHNOLOGY AND MODIFY YOUR DECISION-MAKING PROCESS BASED ON THE MOST CURRENT STATE

Adaptability and flexibility as voice technology advances are going to be important for making ethical decisions that are in everyone’s best interests. The Open Voice Network will continue to monitor and update our guidance, and we hope practitioners will continue to monitor these guidelines and provide feedback as well.

7.4 CONTINUOUSLY AUDIT VOICE-SPECIFIC DECISIONS AND PROCESSES AND KEEP TRACK OF YOUR PROGRESS OVER A PERIOD OF TIME

Auditability of voice experiences over time will inform its creators on how they are doing, if they are progressing, and what changes could be needed to shape a trajectory of trustworthy voice experience design and development.

7.5 GET INVOLVED WITH THE OPEN VOICE NETWORK

Support the development of the Open Voice Network TrustMark Program
Share 30 minutes and tell us about the ethical issues you and your customers are facing, what you’re doing about it, and what the Open Voice Network should do about it
Become an ethical use reviewer and advisor
Become a sponsor
Provide feedback on the work that we are doing
Share our work with others to amplify our message
Attend our events
Join our committees and workgroups
Reach out to collaborate with us

8.0 ABOUT THE OPEN VOICE NETWORK

The Open Voice Network (OVON) is a non-profit industry association dedicated to the development of standards for voice assistance transparency, consent, limited collection, and control of voice data that will make using voice technology worthy of user trust. In any reality, virtual or otherwise, we believe personal privacy should be respected as the default. The Open Voice Network operates as an open-source community within The Linux Foundation. It is independently funded and governed with participation from more than 120 voice practitioners and enterprise leaders from 12 countries. The Open Voice Network community’s work is open source. We seek inclusive input and like to share our insights. At present, our work is focused in four areas:

Interoperability, defined as the ability for conversational agents to share dialogs (and accompanying context, control, and privacy)
Destination registration and management, the ability of users to confidently find a destination of choice through specific requests, and for the providers of goods and services to register a verbal “brand”—similar to the Domain Name System (DNS) of the internet;
Privacy, with voice-specific guidance for both the protection of individual user data and that of commercial users; and
Security, with a focus on voice-specific threats and harms.

Please see our papers and support the Open Voice Network by visiting openvoicenetwork.org.

9.0 ABOUT THE LINUX FOUNDATION

Founded in 2000, The Linux Foundation is supported by more than 1,000 members and is the world’s leading home for collaboration on open-source software, open standards, open data, and open hardware. Linux Foundation’s projects are critical to the world’s infrastructure including Linux, Kubernetes, Node.js, and more. The Linux Foundation’s methodology focuses on leveraging best practices and addressing the needs of contributors, users, and solution providers to create sustainable models for open collaboration. For more information, please visit us at linuxfoundation.org.

10.0 ACKNOWLEDGEMENTS

Authored by the Open Voice Network with special thanks to Ethical Use Task Force contributors Audrey Amsellem, Audrey Arbeeny, Dr. Maria Aretoulaki, Valeria Barbero, Emily Banzhaf (document draft), Leigh Clark, Oita Coleman, David Devoto, Brooke Hawkins, Tom Hewitson, John T. LaGrand, Janice Mandel, Shruti Murali, Michael Novak, Sara Oliver, Leslie Pound, Emerson Sklar, Jon Stine, Emily Troutner, and Nigel Write; edited by Janice Mandel.

11.0 REFERENCE LIST

American Bar Association. (2016, May 20). “Who am I talking to?” - The regulation of voice data collected by connected consumer products. Business Law Today. https://www.american bar.org/groups/business_law/publications/blt/2016/05/06_black/

Ammari, T., Kaye, J., Tsai, J. Y., & Bentley, F. (2019). Music, search, & IoT: How people (really) use voice assistants. ACM Transactions on Computer-Human Interaction, 26(3), 17-1. https://doi.org/10.1145/3311956

Ard, A. J. (2021, December 3). Amazon, can we have our name back?. Washington Post. https://www.washingtonpost.com/technology/interactive/2021/people-named-alexa- name-change-amazon/?itid=sf_technology_article_list

Barrett, B. (2019, August 2). Hey, Apple! Opt out is useless. Let people opt in. Wired. https://www.wired.com/story/hey-apple-opt-out-is-useless/

Baugh, J. (2002). Linguistic Profiling. In S. Makoni, G. Smitherman, A. F. Ball, & A.K. Spears (Eds.), Black linguistics: Language, society, and politics in Africa and the Americas (1st ed., pp. 155-168). Routledge. https://doi.org/10.4324/9780203986615

Behr, M. (2020, December 17). Meet Sam: The non-binary voice for a digital assistant. Digit. https://www.digit.fyi/meet-sam-the-non-binary-voice-for-a-digital-assistant/

Bheemaiah, K., & Esposito, M. (2021). AI ain’t cheap. Capgemini. https://www.capgemini.com/ 2021/01/ai-aint-cheap/

Burns, V., (2021, April 2). How voice recognition is helping HR save time and hire smarter. Unleash. https://www.unleash.ai/how-voice-recognition-is-helping-hr-save-time-and- hire-smarter/

Byrum, J. (2020, May 18). Build a diverse team to solve the AI riddle. MIT Sloan Management Review. https://sloanreview.mit.edu/article/build-a-diverse-team-to-solve-the-ai-riddle/

Carr, A., Day, M., Frier, S., & Gurman, M. (2019, December 11). Silicon Valley is listening to your most intimate moments. Bloomberg Businessweek. https://www.bloomberg.com/news/ features/2019-12-11/silicon-valley-got-millions-to-let-siri-and-alexa-listen-in

Chen, A. (2019, March 14). Why companies want to mine the secrets in your voice. The Verge. https://www.theverge.com/2019/3/14/18264458/voice-technology-speech-analysis-ment al-health-risk-privacy

Chin, C., & Robison, M. (2020, November 23). How AI bots and voice assistants reinforce gender bias. Brookings Institute. https://www.brookings.edu/research/how-ai-bots-and-voice- assistants-reinforce-gender-bias/

Clark, L., Cowan, B. R., Roper, A., Lindsay, S., & Sheers, O. (2020). Speech diversity and speech interfaces: Considering an inclusive future through stammering. Proceedings of the 2nd Conference on Conversational User Interfaces (CUI ‘20). Association for Computing Machinery, New York, NY, USA, Article 24, 1-3. https://doi.org/10.1145/3405755. 3406139

Claypoole, T. F. (2021). Voice analysis complicates personal privacy. National Law Review, XI(271). https://www.natlawreview.com/article/voice-analysis-complicates- personal-privacy

Cohen, J. (2021, October 22). Amazon’s Alexa collects more of your data than any other smart assistant. PC Mag. https://www.pcmag.com/news/amazons-alexa-collects-more-of-your- data-than-any-other-smart-assistant

Crawford, K., & Joler, V. (2018, September 7). Anatomy of an AI System: The Amazon Echo as an anatomical map of human labor, data, and planetary resources. AI Now Institute and Share Lab. https://anatomyof.ai/

Day, M., Turner, G., Drozdiak, N. (2019, April 10). Amazon workers are listening to what you tell Alexa. Bloomberg. https://www.bloomberg.com/news/articles/2019-04-10/is-anyone- listening-to-you-on-alexa-a-global-team-reviews-audio

Despotovic, V., Ismael, M., Cornil, M., Mc Call, R., & Fagherazzi, G., (2021). Detection of COVID-19 from voice, cough, and breathing patterns. Computers in Biology and Medicine, 138, 1-9. https://doi.org/10.1016/j.compbiomed.2021.104944

Fagherazzi, G., Fischer, A., Ismael, M., & Despotovic, V., (2021). Voice for health: The use of vocal biomarkers from research to clinical practice. Digital Biomarkers, 5(1), 78-88. https://doi.org/10.1159/000515346

Gershgorn, D. (2019, September 26). Why it’s so hard for Amazon Alexa to really explain itself. Medium. https://onezero.medium.com/why-its-so-hard-for-amazon-alexa-to-really- explain-itself-9d5ede67a68

Graham, J. (2017, July 19). ‘Alexa, call 911’ won’t work. Here’s what will. USA Today. https://www.usatoday.com/story/tech/talkingtech/2017/07/19/alexa-cant-dial-911-but-g oogle-alexa-and-siri-can-get-you-help/486075001/

Haque, A. B., Shurid, S., Juha, A. T., Sadique, M. S., & Asaduzzaman, A. S. M. (2020, February 2-5). A novel design of gesture and voice controlled solar-powered smart wheel chair with obstacle detection [Paper, pp. 23-28]. IEEE International Conference on Informatics, IoT, and Enabling Technologies (ICIoT), Doha, Qatar. https://doi.org/10.1109/ICIoT48696. 2020.9089652

IEEE. (2019). Ethically Aligned Design (1st ed.). https://standards.ieee.org/wp-content/uploads/ import/documents/other/ead1e.pdf

Ingale, A. B., & Chaudhari, D. S. (2012). Speech Emotion Recognition. International Journal of Soft Computing and Engineering (IJSCE), 2(1). 235-238.

Intelligence Community. (2020). Artificial intelligence ethics framework for the intelligence community. https://www.intelligence.gov/artificial-intelligence-ethics-framework- for-the-intelligence-community#Review

Jiřík, P. (2021, October 19). Voice Biometrics in the Banking Industry. Phonexia. https:// www.phonexia.com/blog/voice-biometrics-in-the-banking-industry/

Kennedy, J., & Strangers, Y. (2020). The smart wife: Why Siri, Alexa, and other smart home devices need a feminist reboot. MIT Press.

Kerkeni, L., Serrestou, Y., Mbarki, M., Raoof, K., Ali Mahjoub, M., & Cleder, C. (2020). Automatic Speech Emotion Recognition Using Machine Learning. In A. Cano, Social Media and Machine Learning. IntechOpen. https://doi.org/10.5772/intechopen.84856

Koenecke, A., Nam, A., Lake, E., Nudell, J., Quartey, M., Mengesha, Z., Toups, C., Rickford, J. R., Jurafsky, D., & Goel, S. (2020). Racial disparities in automated speech recognition. Proceedings of the National Academy of Sciences (PNAS), 117(14), 7684-7689. https://doi.org/10.1073/pnas.1915768117

Laver, J. (1980). The phonetic description of voice quality: Cambridge studies in linguistics. Cambridge University Press.

Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50-80. https://doi.org/10.1518/hfes.46.1.50_30392

Lerman, R. (2021, September 2). Lawsuits say Siri and Google are listening, even when they’re not supposed to. Washington Post. https://www.washingtonpost.com/technology/2021/ 09/02/apple-siri-lawsuit-privacy/

Li, M. (2020, October 26). To build less-biased AI, hire a more diverse team. Harvard Business Review. https://hbr.org/2020/10/to-build-less-biased-ai-hire-a-more-diverse-team

Maghilnan, S., & Kumar, M. R. (2017, June 23-24). Sentiment analysis on speaker specific speech data [Paper, pp. 1-5]. 2017 International Conference on Intelligent Computing and Control (I2C2), Coimbatore, India. https://doi.org/10.1109/I2C2.2017.8321795

Mayeux, R. (2004). Biomarkers: Potential uses and limitations. NeuroRX: the Journal of the American Society for Experimental NeuroTherapeutics, 1(2), 182-188. https://doi.org/ 10.1602/neurorx.1.2.182

McMillan, G. (2011, June 1). It’s not you, it’s it: Voice recognition doesn’t recognize women. Time. https://techland.time.com/2011/06/01/its-not-you-its-it-voice-recognition- doesnt-recognize-women/

Murali, Shruti. (2022, October 11). Designing for Emergency Situations [Conference session]. VOICE 22, YouTube. https://www.youtube.com/watch?v=MVC1ai5gDpo

Newman, L.H. (2019, September 23). Google tightens its voice assistant rules amid privacy backlash. Wired. https://www.wired.com/story/google-assistant-human-transcription- privacy/

Nieva, R., & Rubin, B. F. (2019, September 25). Amazon Alexa adds new commands to tamp down privacy concerns. CNET. https://www.cnet.com/home/smart-home/amazon-alexa- adds-new-commands-to-tamp-down-privacy-concerns/

Open Voice Network. (2022a, October 20). Data Security Specific to Voice. https://openvoice network.org/docs/data-security-specific-to-voice/

Open Voice Network. (2022b, September 22). Interoperability of Conversational Assistants. https://openvoicenetwork.org/docs/interoperability-of-conversational-assistants/

Open Voice Network. (2023, March). Privacy Principles and Capabilities Unique to Voice. https://openvoicenetwork.org/docs/privacy-principles-and-capabilities-unique-to-voice/

Orduña, N. (2019, July 16). AI-driven companies need to be more diverse. Here’s why. World Economic Forum. https://www.weforum.org/agenda/2019/07/ai-driven-companies- need-to-be-more-diverse-here-s-why/

Pinto, R. (2021, July 28). Voice authentication: How it works & is it secure?. 1Kosmos. https://www.1kosmos.com/biometric-authentication/voice-authentication/

Rice, P. (2006, February 2). Linguistic profiling: The sound of your voice may determine whether you get that apartment or not. The Source, Washington University in St. Louis. https://source.wustl.edu/2006/02/linguistic-profiling-the-sound-of-your-voice-may-dete rmine-if-you-get-that-apartment-or-not/

Robison, M. (2020, December 9). Voice assistants have a gender bias problem. What can we do about it?. Brookings Institute. https://www.brookings.edu/blog/techtank/2020/12/09/ voice-assistants-have-a-gender-bias-problem-what-can-we-do-about-it/

Schönherr, L., Golla, M., Eisenhofer, T., Wiele, J., Kolossa, D., & Holz, T. (2020, August 2). “Unacceptable, where is my privacy?” Exploring accidental triggers of smart speakers. arXiv. https://arxiv.org/abs/2008.00508

Schwartz, E. H. (2021, November 9). Google tests voice assistant for people with speech impairments. Voicebot.ai. https://voicebot.ai/2021/11/09/google-tests-voice-assistant- for-people-with-speech-impairments/

Securiti. (2022, August 29). Opt In vs Opt Out Consent: What’s the Difference?. https://securiti.ai /blog/opt-in-vs-opt-out/

Shastri, A. (2020, July 1). Diverse teams build better AI. Here’s why. Forbes. https://www.forbes. com/sites/arunshastri/2020/07/01/diverse-teams-build-better-ai-heres-why/?sh=662f3e 4177b3

Sivakumar, R. (2020, October 13). Voice Biometric Authentication Explained. Claysys. https:// www.claysys.com/blog/voice-biometric-authentication/

Snow, J. (2019, January 30). How people with disabilities are using AI to improve their lives. PBS. https://www.pbs.org/wgbh/nova/article/people-with-disabilities-use-ai-to-improve-their-lives/

The Editorial Board. (2021, March 6). America, Your Privacy Settings Are All Wrong. The New York Times. https://www.nytimes.com/2021/03/06/opinion/data-tech-privacy-opt-in.html

TWB Communications. (2020, September 1). TWB and KoBo Inc develop speech recognition technology to capture voices of speakers of marginalized languages. Translators Without Borders. https://translatorswithoutborders.org/twb-and-kobo-inc-develop-speech- recognition-technology-to-capture-voices-of-speakers-of-marginalized-languages/

UNESCO. (2019). I’d blush if I could: Closing gender divides in digital skills through education. https://en.unesco.org/Id-blush-if-I-could

UNI Global Union. (2017). Top 10 principles for ethical artificial intelligence. http://www.the futureworldofwork.org/media/35420/uni_ethical_ai.pdf

United Nations. (1948). Universal Declaration of Human Rights. https://www.un.org/sites/un2. un.org/files/udhr.pdf

United Nations General Assembly. (2015). Transforming our world: The 2030 agenda for sustainable development. https://sdgs.un.org/2030agenda

Vixen Labs. (2022). Voice Consumer Index 2022. https://vixenlabs.co/wp-content/uploads/ 2022/06/VixenLabs_VoiceConsumerIndex2022.pdf

Wong, K. (2019, October 1). How to be a more conscious consumer, even if you’re on a budget. New York Times. https://www.nytimes.com/2019/10/01/smarter-living/sustainabile- shopping-conscious-consumer.html

Wroge, T. J., Ӧzkanca, Y., Demiroglu, C., Si, D., Atkins, D. C., & Ghomi, R. H. (2018, December 1). Parkinson’s disease diagnosis using machine learning and voice [Paper, pp. 1-7]. IEEE Signal Processing in Medicine and Biology Symposium, Philadelphia, Pennsylvania, USA. https://doi.org/10.1109/SPMB.2018.8615607

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Files

AI_Ethical_Guidelines_For_Voice_Experiences.md

Latest commit

History