We build innovative AI algorithms to help measure mental health. Because we are entrusted with very sensitive, personal data, we prioritize data security and AI fairness. We are committed to researching and refining our technology with diverse groups of people to ensure our tools can support earlier detection and better outcomes for all. There is significant research suggesting that when someone is experiencing depression, anxiety, suicidality or mental distress, their speech patterns change without them realizing it. To measure this, we have developed Clairity Artificial Intelligence (Clairity AI), our proprietary model that analyzes speech to help measure mental health risk. Our model is based on more than 10 years of clinical research on natural language processing (NLP) and machine learning (ML) algorithms. This research showed that there are suicidal “thought markers” that cause changes in speech, which can be detected and measured with AI. Our AI models analyze client/patient speech to identify vocal biomarkers for mental health risk. To do this, voice samples are securely recorded, transmitted to a HIPAA compliant cloud server, and analyzed by our patented AI models. Because we are entrusted with very sensitive, personal data, we prioritize data security and ensure data remains encrypted in transit and at rest. Clarigent Health’s mission is to equip mental health professionals and their supervisors with state-of-the-art technology, so they can perform at their highest levels by providing better data, novel analysis, and intelligent insights.

Mental health screenings allow for early identification and intervention and help bridge the gap. Resulting with improvement in quality of life and reduction of cost to both the system and to families. Depression and Anxiety are some of the most common mental health diagnoses detected from mental health screenings, both of which are strongly correlated to risk of suicidality. Mental health illnesses are real, common, and treatable; recovery is very possible with the right steps, especially if detected early. Participating in a mental health screening is just as important as getting regular check-ups with your doctor. Seven to ten million people in the United States have at least one mental health disorder, in addition to an existing alcohol or drug abuse disorder. Too often, these individuals are treated for only one of the two disorders or receive no treatment at all. In an effort to bridge the gap, mental health screenings help address these discrepancies and allow people to receive not only the help they need, but the help they deserve.

Clairity is a person-centered mental health screening approach for anxiety, depression, and suicidality. Mental Health Providers and Care Teams are constantly faced with making decisions. Does someone need care? If so, at what level of service? What is the urgency? What is the crisis level? Clairity is a clinical decision support tool for mental health professionals and care teams can be utilized as a pre-screen, between the visit, in-visit or as a step-down touchpoint to identify risk and inform referrals, treatment, and diagnosis. The Clairity approach as less accusatory, empathetic, encourages active listening, and validating of an individual’s core emotions.

Clairity is a decision support tool intended for use by mental health professionals and care teams to inform mental health risks including anxiety, depression and suicidality. Because Clairity is technology advanced and automated using artificial intelligence, it can be quickly deployed in various settings and in different intercept care points, and the results are more portable and can be integrated into health record and care plans. Clairity has been implemented in Emergency Departments to inform behavioral health triage, justice systems to inform behavioral health referrals at pre-adjudication, outpatient intake to inform diagnosis and treatment plans, between visit check-in to monitor between therapy visits, and intake for digital veteran support to develop risk profile.

Clairity is a decision support tool intended for use by mental health professionals and care teams to inform mental health rsks including anxiety, depression and suicidality. Access to use Clairity is purchased with a subscription for using the service. Contact Clarigent Health for more information at info@clarigenthealth.com

Using a smartphone or Windows computer, Clairity captures linguistic and acoustic data that can be used to classify the individual's mental health risks including anxiety, depression and suicidality. According to research, individuals who are experiencing anxiety, depression, and/or suicidality change their thought patterns, causing their speech patterns to change as well. The changes in speech are subtle, and usually cannot be detected in a conversation. However, researchers have developed models that count and calculate the changes. Clairity AI builds on this research and applies machine learning to make ever more accurate predictions. These thought markers reflect underlying changes in brain chemistry making them ideal candidates for early risk identification.

A vocal biomarker is a signature, a feature, or a combination of features from the audio signal of the voice that is associated with a clinical outcome and can be used to monitor patients, diagnose a condition, or grade the severity or the stages of a disease or for drug development Clairity analyzes and transforms an individual’s speech into vocal biomarkers to identify mental health risks, including anxiety, depression, and suicidality. Nuances in voice and language patterns reveal thought markers related to emotional suffering. These nuances can be detected and measured using Clairity’s Artificial Intelligence technology.

Clarigent Health recognizes that the information we collect about individuals during screening and therapy sessions is highly sensitive. We ensure that this data is encrypted any time it is uploaded or downloaded from our HIPAA compliant servers. We do not allow it to be stored on interviewers’ or therapists’ devices. Rather, the data is stored in our secure cloud environment that requires password authenticated access. We never share protected health information (PHI) with anyone except the provider caring for the individual, and the supervisors who oversee the provider. The Clairity report has two levels of permission depending on user: Therapist or Supervisor. The “Therapist” will be able to view only their own clients or patients. A “Supervisor” will have access to all clients belonging to the Therapists that report to them. Data access will be restricted by login credentials and optional Multi-Factor Authentication, with appropriate permission applied for each user as designated by their organization.

No, Clairity is administered by a mental health professional or care team member.

Currently, Clairity AI analyzes linguistics, or the words spoken by individuals who speak English as a primary language.

Reimbursement codes aligning with Clairity Code 96136 Current Procedural Terminology (CPT) coding has been established for psychological and/or neuropsychological assessments and is contained in the CPT manual within section, “Central Nervous System Assessments/Tests (e.g., Neuro-Cognitive, Mental Status, Speech Testing)”. Notably, in the instructions for this section of codes, there is an expectation that the administration of the test(s) characterized by the applicable code(s) “will generate material that will be formulated into a report.” Clarigent has developed a mental health Clairity Assessment tool using a person-centered interview administered by a technician or clinician using an electronic platform. The Clairity interview, feeds two tests, returning suicidal and depression risk scores. The Clairity session can be appropriately described by the 96136 code for “Psychological or neuropsychological test administration and scoring by physician or other qualified healthcare professional, two or more tests, any method, first 30 minutes.” The average Medicare allowable fee is $48. The 96136 code requires a minimum of 16 minutes to be billed. The Clairity interview is brief, requiring 5-10 minutes. However, the entire length of the Clairity session including the input into the app of mental health status ratings (based on the DSM-V) and receipt of scores should be considered. Standard procedure for patient evaluation may also include other assessments (e.g., PHQ-9) along with the Clarigent assessment, which would count toward the time as well. As with all claims, relevant diagnosis coding (ICD-10) is an important inclusion to convey the reason the testing was performed. For patients with mental health benefits and receiving services from a behavioral health provider, the F-series ICD-10 codes are typically best to justify the related testing. Claims are unlikely to be accepted for “routine re-evaluation of chronically disabled patients that is not required for a diagnosis or continued treatment,” so it is important to highlight how the Clairity assessment is informing the treatment plan and ongoing decision making and information sharing with the client. Code 96138 The 96138 code can also be used for “Psychological or neuropsychological test administration and scoring by technician or other qualified healthcare professional, two or more tests, any method, first 30 minutes.” The requirements are similar to 96136, but the licensure of the administrating professional are different. The average Medicare allowable fee is $38. Modifier 59 Code 96138 or 96136 may be billed as a standalone code for the Clairity procedure conducted and reviewed between regular therapy sessions. However, if it is billed as an add-on to the regular session which its results inform, it may be billed alongside the usual therapy codes using modifier 59. Code 96127 Code 96127 could be billed in place of 96136 or 96138. However, the average Medicare allowable fee is $4.89 per assessment. In the case of Clairity, two assessments (Depression, Suicide risk) could be claimed as two units.

Clairity detects signals that are likely nonconscious - meaning that the individual is often unaware they are sending these signals. This makes it difficult for the individual to distort their own score or "game" the system. Clairity is making a complex calculation that humans cannot easily do in their heads in real-time. This means that the signal may be missed by humans, even highly trained and experienced therapists.

If the individual is nonverbal, or chooses to remain silent, Clairity has no speech sample to analyze. Audio analysis cannot record body language, facial expression, or physical appearance/hygiene. Clairity is currently focused on linguistic analysis and will likely incorporate acoustic and other aspects in the future.

Despite continued efforts to reduce suicide rates, they have steadily increased for over a decade. Suicide is the 10th leading cause of death in the US In 2019, 47,51 Americans died by suicide In 2019, there were an estimated 1.38M suicide attempts The age-adjusted suicide rate in 2019 was 13.93 per 100,000 individuals. The rate of suicide is highest in middle-aged white men. In 2019, men died by suicide 3.63x as often as women. On average, there are 130 suicides per day. White males accounted for 69.38% of suicide deaths in 2019. In 2019, firearms accounted for 50.39% of all suicide deaths. Nearly 90% of all suicides are associated with a diagnosable mental health or substance-abuse disorder. 93% of adults surveyed in the U.S. think suicide can be prevented. Mental health clinics, schools, hospitals and other settings, consistent screening of suicide risk is underutilized, and many providers are uncertain about the value of universal screening approach, and often feel underprepared to address suicide thoughts and behaviors (STBs) when they are present. The idea behind universal screening is that although suicide death rates are low, suicidal thoughts, behaviors, and attempts occur at exceptionally higher rates. Interrupting STBs before they advance to attempts or death is best done by broad and regular screenings.

Screening is important to identify individuals who may be at risk for mental health issues including anxiety, depression, and suicidality. Assessment is a more comprehensive evaluation done by a clinician to confirm suspected risk, estimate the immediate danger, and decide on course of treatment. Clairity is a qualitative objective screening for mental health risks.

A mobile (iOS or Android) and Windows app used to capture amental health interview and clinical impressions for a real-time mental health risk scores including anxiety, depression and suicidality.

A semi-structured interview designed to provoke positive and negative emotional responses related to what experts consider key contributions to suicidality. Clients/patients should be screened for suicidal risk on their first contact and at every subsequent contact if the risk exists. Clairity is an objective consistent screening to identify and monitor individuals for risk of mental health crisis including suicidality. These questions address emotional states and aim to allow the client/patient to freely disclose their feelings rather than require responses of more specific questions typically asked in a mental health assessment scale. People report how they feel using words. In everyday life, when we want to know how someone is feeling, we ask. “How are you feeling?” is one of the most ubiquitous questions in our culture. Psychologists have relied on self-reports to find out how people feel for experienced emotion (e.g., anger, sadness, fear). There is no known objective, external measure of the subjective, internal events that we experience as anger, sadness, fear, and so on. If we want to know whether people feel these emotions, we have to ask them. The Clairity semi-structured interview draws out a person’s emotions by talking about their feelings (probing in areas of Hope, Secrets, Anger, Fear and Emotional pain), and can distinguish suicidal risk from speech responses in a 5-minute interview. Clairity is used to promote conversation where thought markers will be expressed. The interview is not yes/no queries; rather they are prompts to start a discussion. They are not diagnostic or prescriptive; they are a semi-structured way to harvest a patient language to analyze speech patterns using the Clairity AI model.

The Clairity semi-structured interview draws out a person’s emotions by talking about their feelings (probing in areas of Hope, Secrets, Anger, Fear and Emotional pain), and can distinguish suicidal risk from speech responses in a 5-minute interview. 1. Hope - Do you have hope and how does that feel? (The most dangerous form of suicidal desire is caused by the simultaneous presence of two interpersonal constructs—thwarted belongingness and perceived burdensomeness, and hopelessness about these states. Consensus in the field that hopelessness is an important predictor of suicidality.) 2. Secrets - Do you keep secrets and how does that feel? (Deadly secrets have a strong association to suicidal risk.) 3. Anger - Do you have anger and how does that feel? (Tends to reveal constructs related to revenge/ blame/ perceived hurt by others.) 4. Fear - Do you have fears and how does that feel? (Greater fears of negative evaluation were significantly and positively associated with levels of suicidal ideation.) 5. Emotional Pain - Do you experience emotional pain and how does that feel? (failure/ burden to others (interesting connection to Joiner’s constructs) and associations between suicide and familial discord, domestic violence, familial stress, and perceptions that one is a burden on family.)

Windows – Go to https://clairity.ai/download to download iOS – Update Clairity from your App Store Android – Update from your Google Play Store Amazon – Update from Amazon Store NOTE - We do not support Mac OS

You must provide permission for the device running the Clairity Capture app to use the microphone. iOS - Clairity asks for microphone permission when starting session Android – You must specifically provide permission to use the microphone when using Clairity. Go to your phone’s settings > AppsClick on Clarigent Clairity > Enable permissions Windows - Ensure your microphone is not disabled

No, not at this time. If you are conducting a virtual interview on a Mac, you can place your mobile device running Clairity next to your Mac to capture the session. Make sure your sound is turned up on your Mac and the mobile device is placed near your speaker.

After authenticating into Clairity Capture, you will be on the [Subject] page. From the bottom navigation menu, click Add [subject].

You can add a single client at a time in the Clarity Capture app. If you would like to bulk load a set of clients, please contact Clarigent Support at support@clarigenthealth.com.

These prompts were originally developed by John Pestian, PhD. with consultation from Ed Schneidman, PhD, considered the "father of suicidology". John Pestian’s bioinformatics lab at CCHMC spent over 10+ years researching suicide and its characteristics using artificial intelligence and machine learning technologies to collect and analyze a person’s language to detect warning signs of distress and suicidal ideation. John and his team began this research having accumulated the largest corpus of suicide notes to understand the language of suicide. These findings were then applied in 3 clinical trials to identify vocal biomarkers for assessing suicidal risk, resulting in 5 patents.

The Clairity Radar data dashboard uses Microsoft Power BI to provide you access to your data in our resources. Clairity provides a data dashboard to the clinician and care team highlighting which clients should be on their radar along with comprehensive reports on each client with key insights derived from vocal analysis from the mental health interview to guide diagnosis and treatment.

By clicking one of the Mental Health risk buttons (anxiety, depression, suicidality), the Radar page will be filtered to report back on individual's that are at risk for selected state as determined by either the Clairity AI score, interviewer's impressions as collected in the Clairity Capture app, or the appropriate patient-rated scale if administered.

This page shows you how one client’s symptoms are progressing from session to session over time. The client is identified in the Client Selector in the top right. Just next to their name, their diagnosis is displayed, which is pulled from their electronic health record. Based on their diagnosis, they may have trends displayed for Overall Distress plus suicidal risk, depressive symptoms and/or anxiety symptoms if relevant. For Overall Distress, you will see one (purple) trendline, based on observed data inputted by the therapist. For Suicidal Risk and presentation of depression and anxiety, you may see up to three trendlines: Blue trendline - based on observations from the therapist (mental health states rated 0-5 in the Clairity app) Purple trendline - based on the Clairity AI score from the client’s language Green trendline – based on the client’s response to the relevant assessment (such as the Columbia scale, PHQ-9, or GAD-7) You may not see all three trendlines, which may mean the data is unavailable or the mental health state is not relevant for this client. If a line is not continuous, it could mean that data is unavailable or missing for some session dates. The bottom axis shows the dates of the sessions. The date filter at the top allows you to select a different time window. To view a detailed summary for a specific session, select the date of interest from the filter at the top and then click the “Session Summary” button.

The PHQ-9 (Patient Health Questionnaire – 9, or PHQ9-A modified for adolescents) is a commonly used module that screens for the presence and severity of depression. It asks the patient questions about their symptoms relating to the nine DSM-IV criteria for depression.  The patient is asked to rate their symptoms from the past two weeks on a scale of “0” (not at all) to “3” (nearly every day). If the patient reports symptoms from four of the nine categories with a total score above 10 (adults) or 11 (adolescents), a depressive disorder is considered. The total score on the questionnaire is also calculated to determine severity of symptoms. The PHQ-9 has been adapted for various groups, including adolescents (PHQ-A), which includes a total of three suicide-related questions. Because of these additional questions related to suicide, we use the PHQ-A instead of the PHQ-9.  

The GAD-7 is a standard 7-item questionnaire that asks the patient questions about their symptoms relating to the DSM-IV criteria for anxiety.  The patient is asked to rate their symptoms from the past two weeks on a scale of “0” (not at all) to “3” (nearly every day). The total scores can range from 0-21, with a total score of 10 (adults) or 8 (adolescents) often used to identify anxiety.

The Columbia scale (C-SSRS short form) is a standard questionnaire with five probes relating to suicidal ideation. The scale returns a suicide risk score of None, Low, Medium, and High. Depending how the patient answers, the test administer may follow up regarding the intensity of the feelings. It is always critical to also consider any behavior that is indicative of suicidal risk. For the purposes of Clairity, any level above “None” is reported as a risk. 

Data we capture, including audio from the therapy session, is securely uploaded to our HIPAA compliant servers. This data is then processed using Clarigent’s patented natural language processing and machine-learned algorithm, Clairity AI. Following processing, the data is returned in an easy-to-digest Clairity report. Our goal is to make this report available in near real-time. However, we also want the report to be complete and to sync with other record tools you use. Depending on our agreement with your organization, it may take up to two business days for us to manage appropriate security checks and pull the data into the report.

“Notes” are pulled directly from the Clairity app – entered by the therapists either by typing or voice-to-text. Clairity auto-generates smart phrases based on data collected in the Clairity app to help cue the therapist’s memory and enrich clinical notes.

The Assessments page displays results of the questionnaires rated by the client. There may be up to three types of assessment depending on the client’s diagnosis: standard scales for depression, suicidal risk, and anxiety.

The client is identified in the Client Selector in the top right corner of the page.

To view the data for a different session, change the date in the filter at the top right of the page

The Clairity report has two levels of permission depending on user: Therapist or Supervisor. The “Therapist” will be able to view only their own clients or patients. A “Supervisor” will have access to all clients belonging to the Therapists that report to them. Data access will be restricted by login credentials, with appropriate permission applied for each user as designated by their organization. Based on your level of access, you can filter by organization, by therapist, or by individual names.

This section allows the user to tab between four types of word clouds. In the first tab (“Most Common”), the size of the word denotes the frequency of the word used by the client. Stop words, common words that do not contribute much to the meaning of a sentence or phrase (such as ‘a’, ‘an’, ‘in’), are removed from word clouds. “Most Common” considers all other words spoken, aside from these, and shows the user which words were used most frequently. The second tab shows the “Most Impactful” word cloud. This word cloud looks at the language that was most important for the Clairity AI model, and displays the words that most contributed to the suicide risk score. This allows the user some insight into what words triggered the risk classification. The third tab shows the “Positive” word cloud. This word cloud only considers the words the client spoke that had a positive valence. The size of the word reflects how often that word was spoken. This allows the user to see what positive topics came up frequently for the client. The fourth tab shows the “Negative” word cloud. This only considers the negative valence words spoken by the client. The size of the word reflects how frequently the word was spoken. This allows the user to quickly see some of the negative topics that were discussed.

Sentiment analysis, like positive and negative valence words, attempts to classify the polarity of language by how positive, neutral, or negative it is. The sentiment analysis provided here uses a tool called The Valence Aware Dictionary for sentiment Reasoning (VADER). It is constructed using word banks similar to the one developed by Charles E. Osgood, but also includes slang and other types of inputs. Click to learn more about the research behind VADER.  Every sentence the client speaks is analyzed with VADER to determine that sentence’s sentiment, ranging from “Very Negative” to “Very Positive”. The Overall sentiment is an average over all the client’s spoken sentences. The Most Positive and Negative Sentences display sentences categorized at the extremes of rated sentiment. 

Basic statistics are provided to show how much of the time the client was speaking, and how much the therapist was speaking. All other measurements are focused solely on the client’s speech.

The interviewer provides ratings in the Clairity Capture app, based on their impression in the session. These ratings power the interviewer impression sections of the report. Clairity also auto-generates smart phrases based on data collected in the Clairity app to help cue the care team's memory and enrich clinical notes. These smart phrases can be found on the right side of the Session Summary page.

We can provide the full transcript upon request. Please send an email to support@clarigenthealth.com.

This means the score is not yet available, but is being processed. The audio captured during the interview is sent in 1-minute segments up to our secure HIPAA compliant servers. This data is then processed using Clarigent’s patented natural language processing and machine-learned algorithm, Clairity AI. The risk scores are returned within minutes using our proprietary Automatic Speech Recognition (ASR) process.

The interviewer provides data in the Clairity app, based on their actions taken in the session.

The suicidal risk level is based on the vocal biomarkers calculated by Clairity AI, which identifies up to 4000 linguistic features in the client’s speech. Based on the linguistic features, the speech sample from the client is determined to be in one of three bins: Undetected, Moderate, or High risk. These bin classifications were developed building on more than ten years of research led by Dr. John Pestian at Cincinnati Children’s Hospital.

Clairity AI is a machine-learned model that analyzes the client’s speech during therapy sessions and predicts suicide risk based on our proprietary algorithms

Natural language processing (NLP) is a subfield of artificial intelligence that uses computers to identify patterns in human language. To process natural language, computers translate words into numbers, and then use mathematical techniques such as ML to analyze and make predictions. For example, we might design a model to predict whether a movie review is good or bad. To train the model we upload labeled data. In this case, labeled data would be reviews, which we have marked as either good or bad. The scientist builds a model to identify when someone has written a bad review. To do this, the model counts the words that appear for each type of review. The words ‘bad’ and ‘terrible’ appear more frequently in bad reviews, so this helps the computer classify bad reviews. As we collect more data, the computer learns that the word ‘boring’ also tends to mean it is a bad review. It updates the model and is now better at classifying the next round.

While terms like ‘machine learning’ and ‘artificial intelligence’ can sound mystical or complex, the underlying principles are quite simple. In this section we will explain the basics so you can make the most of this technology. Machine learning (ML) uses statistical techniques to find patterns in data. The simplest ML model is y = mx + b(Eq. 1) the equation for a line that many of us learn in seventh or eighth grade. We can train this model with data that we have, to make predictions about things we do not know. In this model (Eq. 1), the variable x is related to y by constants m and b, the slope of the line and the y-intercept, respectively. To train the model, we can use a few coordinate points that we know, which are pairs of x and y. From this, we can solve for the best values of m and b that will fit a line between the points. Once the values for m and b are determined, the model is trained. Now when we show it new values of x, it can predict y. This principle of using training data to design a model that can make predictions remains the same regardless of how mathematically complicated the model becomes. Models can be simple if the relationship between the points is clear, or complex if the pattern is less obvious. When faced with lots of complicated data, data scientists may need a more heavy-duty model than Eq. 1. In these cases, the machine learning scientist can provide a computer with training data and explore models that allow for more elaborate relationships between many variables. As the computer reviews more and more data, it learns by updating its models, much in the way the values of m and b are learned in the example of the equation for a line. Models can be simple if the relationship between the points is clear, or complex if the pattern is less obvious. When faced with lots of complicated data, data scientists may need a more heavy-duty model than Eq. 1. In these cases, the machine learning scientist can provide a computer with training data and explore models that allow for more elaborate relationships between many variables. As the computer reviews more and more data, it learns by updating its models, much in the way the values of m and b are learned in the example of the equation for a line.

Clairity AI is based on years of clinical research at the Cincinnati Children’s Hospital Medical Center. The research was led by Dr. John Pestian, an expert in neuropsychiatric biomedical informatics. Across three clinical trials, Dr. Pestian and colleagues were able to predict with a high degree of accuracy whether a participant was healthy, mentally ill, or suicidal based on ML analysis of their language. In the first clinical trial, Dr. Pestian’s team demonstrated that their machine learning algorithms could identify suicidal language (in a suicide note) with greater accuracy (75%) than a mental health professional alone (63%) (Pestian et all 2010). In the next clinical trial, Dr. Pestian demonstrated that NLP algorithms could recognize if someone was suicidal or nonsuicidal (Pestian et al 2016). In the third study, their improved algorithm could detect not only suicidality, but also whether a patient had mental illness, based on their speech. In this study, participants were assessed by clinicians and standard assessments to determine whether they belonged to one of three classes: no mental health condition or suicidal risk (control), mental health condition and no suicidal risk (mentally ill), or suicidal risk (suicidal). All participants were asked the Ubiquitous Questionnaire (the MHSAFE probes), and the language from their responses to these probes was analyzed. Using a combination of NLP and ML, Dr. Pestian and colleagues improved their algorithms which were able to detect mental illness or suicidality with 85% accuracy (Pestian et al 2017). Clairity AI is based on the algorithms tested in these clinical studies. Clarigent Health continues to invest in research intended to improve the algorithms.

Clairity AI draws from a rich body of research showing how the language we use reflects our mental health and emotional state. Small signals in our speech, including the content words we use and the way we form sentences can give a window into our minds. Because the signals are so subtle, and based on thousands of words we speak, it is very difficult for the human ear to pick up on the patterns. Fortunately, artificial intelligence can help us with this task. Currently, Clairity AI analyzes linguistics, or the words spoken by individuals who speak English as a primary language. Clarigent Health is developing future versions that will incorporate other features such as inflection and intonation.

Currently, Clairity AI analyzes linguistics, or the words spoken by individuals who speak English as a primary language. Clarigent Health is developing future versions that will incorporate other features such as inflection and intonation.