<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Lesa Wright on Medium]]></title>
        <description><![CDATA[Stories by Lesa Wright on Medium]]></description>
        <link>https://medium.com/@lesa-s-wright?source=rss-abecb83e6a47------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*2mfNfY1hFH5O9vIb5iP7mA.jpeg</url>
            <title>Stories by Lesa Wright on Medium</title>
            <link>https://medium.com/@lesa-s-wright?source=rss-abecb83e6a47------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 20:41:30 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@lesa-s-wright/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[From Consultant Psychiatrist to Chief AI Officer: Building High-Impact AI in Mental Health]]></title>
            <link>https://lesa-s-wright.medium.com/from-consultant-psychiatrist-to-chief-ai-officer-building-high-impact-ai-in-mental-health-0f8214381fd0?source=rss-abecb83e6a47------2</link>
            <guid isPermaLink="false">https://medium.com/p/0f8214381fd0</guid>
            <dc:creator><![CDATA[Lesa Wright]]></dc:creator>
            <pubDate>Sat, 20 Dec 2025 19:23:25 GMT</pubDate>
            <atom:updated>2025-12-20T19:23:25.451Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*epoF_fs02cdtAnuX0avYvw.jpeg" /></figure><p><em>I have spent more than twenty years in mental health and over a decade building software to support care. That path has taken me from practising as a Consultant Psychiatrist to leading as a Chief AI Officer.</em></p><p>My last career piece was published in December of 2020 and reflected on my journey over the preceding five years in which I transitioned from being a clinician to being a clinician-technologist. You can read it <a href="https://medium.com/@lesa-s-wright/my-journey-in-the-ai-xprize-dbd5d65bb14e">here</a>.</p><p>Bill Gates’s observation — that people tend to overestimate what can be done in one year and underestimate what can be done in five or ten years — has proved materially true.</p><p>A great deal has happened since.</p><p>This article continues that story for people making significant career transitions into technology and AI leadership, offering a candid view of what that journey can feel like in practice. Through descriptions of my experiences as Chief Technology Officer and then Chief AI Officer, it picks up right where the last article left off, just before the beginning of a surge in demand for ADHD assessment and treatment services.</p><p><strong>Planning for the future and leading from the front.</strong></p><p>Up until that point my “tech team” had consisted of a part-time developer, a project manager/web designer and myself. Earlier that year, I decided to change how our bespoke electronic patient record system was built. We moved away from a setup that used a constant “live” connection between the browser and the server (a web socket) and a database that stores information in flexible, non-table formats (a NoSQL database). Instead, we switched to a more standard approach where the system asks for and receives information through regular web requests (a REST API) and stores data in organised tables with clear links between them (a relational database).</p><p>The original architecture had been slow and difficult to debug and so we looked at simplification, scalability and robustness in that scenario.</p><p>In time, I would introduce a range of practices to support scalability. This would include monitoring and remediating the top 5 system errors and top 5 slow queries every one to two sprints. This allowed us to tackle unpredictable demand in an ever-changing business environment.</p><p>We were in the messy middle — a time in the growth of a business where challenge and opportunity abound in a storm that needs persistent dogged navigation. So, we needed to establish processes that worked for us and would prepare us for the future. Over time, the tech team was grown, and we evolved our development cycle moving from waterfall-like to agile-like practices and rebuilt our platform while “out at sea”.</p><p><strong>A tsunami.</strong></p><p>Over the course of three weeks at the end of 2020, the trajectory of the business changed. I had been with the business just over a year and it had been established seven years before that. When I joined the company, the Psychiatry UK platform had supported less than a thousand active patients each month.</p><p>Psychiatry UK is a pioneer in online telepsychiatry. One of the founders had commented to me that in the early years, Psychiatry UK had consisted of a coalition of the willing. He meant that in the past the established healthcare system regarded remote consultations in psychiatry as quirky rather than mainstream. Only clinicians who really believed in trying something new and committing to make it work joined. This resulted in slow growth in the early years of the business.</p><blockquote>I believe that luck favours fools, but opportunity is a gift for the prepared.</blockquote><p>The people who made up Psychiatry UK invested in time, resilience in the face of doubt and problem solving as preparation for the future.</p><p>The arrival of the COVID-19 pandemic at the beginning of 2020 changed everything. While services around the world scrambled to adapt their mental health care delivery model to online care, Psychiatry UK was already up and running.</p><p>In November 2020, shortly before an unprecedented wave of demand arrived, we launched what we called version 3 of our EHR (Electronic Health Record) platform with a new backend architecture. There were some teething problems that required hot fixes and regular updates to the wider business and within a few days of that transition, the platform had stabilised.</p><p>The wave of referrals arrived in December 2020.</p><p>Week one.</p><p>The first week of the surge opened as usual, I concluded a diagnostic assessment and told my patient that they could start titration onto medication the following week.</p><p>Week two.</p><p>At the beginning of that week, I had to tell different patients that they could only start titration in two weeks’ time.</p><p>Week three.</p><p>At the beginning of the next week, I had to apologise and tell my next patients that it would be four to six weeks before they could start treatment.</p><p>In the weeks that followed, I had to explain to disappointed patients that there had been a surge in assessments, and I was not sure when they would be able to start treatment. When a diagnosis is properly made in someone who meets ADHD diagnostic criteria, it is often like daybreak after a long scary night. Treatment for most enables a better-quality of life in education, work, relationships, and their mental health and for some it is like being set free from a prison of lost dreams and underachievement. Not being able to offer treatment soon after diagnosis is a difficult burden for clinicians, and an even harder one for patients to live with.</p><p>The business began to scale, and the technology team worked relentlessly to support its growth. Over the next two years we would partner with suppliers to help us build and maintain the systems as the once monolithic application became a more modular platform that gave us a competitive edge.</p><p>The business enjoyed a strong reputation in the marketplace and a CQC rating of ‘Good’ — a position that would soon be tested as demand surged beyond all expectations.</p><blockquote>From supporting less than a thousand active patients a month our custom EHR platform would expand and help us serve 70,000 active patients a month in less than four years.</blockquote><p><strong>Anticipating need.</strong></p><p>One key technology that I introduced was changing how patients provided pre-assessment information. We had started out with asking patients to download a Word document that they would complete and then attach to their patient record.</p><p>A simple change was to use a web form to support this instead. There was no suitable privacy-focused off-the-shelf solution at the time. So I built a web form ecosystem with a privacy-centric design so that each form avoided storing personally identifiable information. This was to ensure security in preventing data leakage in analytics and statutory submissions. The web forms also formed the foundation of our treatment monitoring system which in turn enabled the business to build what is likely to be the largest ADHD treatment response dataset in the UK.</p><p>If a diagnosis of ADHD is made, patients are offered medication as treatment if appropriate through titration — which is the process of starting and adjusting medication dosages according to response.</p><p>In the early years of this service, every prescriber either hand wrote prescriptions or used a Word template that they manually edited for each patient before printing it out. Recognising the need for standardised prescription printing to improve efficiency, I designed an in-house solution when commercial e-prescribing options proved too costly.</p><p><strong>Growing the technology team.</strong></p><p>Having learnt my lesson from the AI Xprize, it was important to me that we brought in support to help us deliver the technology we needed. I am grateful to the partners that helped us as the tech demand grew: <a href="https://www.linkedin.com/company/mayantechs/">Mayan Technologies</a>, <a href="https://www.linkedin.com/company/adapt-digital-ltd/">Adapt</a>, <a href="https://www.linkedin.com/company/carefulai-ltd/">CarefulAi</a> and <a href="https://www.linkedin.com/company/insodus-technologies/posts/">Insodus Technologies</a>.</p><p>The team was expanded first with web development contractors. We worked with cybersecurity contractors to improve the security posture of the business. An Information Technology department was formally established to handle company hardware and digital workplace solutions as well as a support desk for patients and staff. We had outsourced much of the IT function in the first instance but found it inadequate for our needs and later brought it in-house.</p><p>Data analytics was introduced to provide business intelligence and management data. A (digital) clinical safety team was formed to prepare for an AI-driven future. The team was tasked with risk assessing technology used for patient care and advising mitigations. The AI team would come later.</p><p>All these components would progress at different rates of maturation. Holding all this together with a tech-wide RASCI (Responsible, Accountable, Supporting, Consulted and Informed) matrix was crucial to understanding the development of different roles and making sense of gaps. A RASCI matrix is a responsibility-assignment tool used in project and organisational management to clarify roles and responsibilities.</p><p><strong>Working with business operations.</strong></p><p>There was an existing “working group” when I joined the business. This was a group of execs and departmental leads that determined business priorities. This was revamped and transformed with formal terms of reference and wider membership. Due to the pressures of rapid growth and limited development capacity, the focus was on patient safety and business-critical requirements. This had always proved challenging and structures around this have evolved dramatically since then.</p><p><strong>AI on the horizon.</strong></p><p>During all this growth, I continued exploring what the role of machine learning and AI would be in the business. I examined the earlier forms of transformers — a type of AI model that’s especially good at understanding and generating text. In particular, I looked at the BERT model (an AI system mainly designed to <em>understand</em> the meaning of written language) and GPT-2 (an AI system mainly designed to <em>produce</em> fluent text). I did this to see whether they could help with the administrative parts of creating clinical letters, such as drafting, structuring, or tidying the wording.</p><p>An ever-present consideration was how any solution would operationalise and how data privacy would be maintained. Due to the nature of mental health assessments which can be upwards of 5 to 10 pages in length, the technology just was not suitable at the time. So, we waited and kept watch, understanding that the technology would only get better.</p><p>I planned the formation of an AI team, considering what impact it should have and how it would achieve that impact. The future looked bright, but with growth came challenge.</p><p><strong>A downgrade.</strong></p><p>In late 2023 the company received a surprise CQC (Care Quality Commission — the UK healthcare provider regulator) inspection. Whilst the business had enjoyed good growth, the CQC was concerned that there were large numbers of patients on waiting lists without adequate safeguards in place.</p><p>The CQC downgraded the business to a rating of “needs improvement” because of this. It was disappointing but they did acknowledge specific areas of <a href="https://api.cqc.org.uk/public/v1/reports/329db31d-38c3-4b40-93ab-7cea4a2ac194?20231122080308">outstanding practice</a> in the business including the role that technology would play in the provision of care once it was ready.</p><p>The CQC noted,</p><blockquote>We found the following outstanding practice:</blockquote><blockquote>The chief technology officer (CTO) and a team of multidisciplinary colleagues were developing a significant, transformational digital programme of work to improve patient care and provider capacity. Work included proof-of-concept testing for clinical artificial intelligence systems at the leading edge of international mental health care development. Technology-led improvements planned for the service took place within clear safety and data protection policies and systems.</blockquote><p>We were executing on our AI plan, but the inspection had taken place too early for us to show its benefits.</p><p>We were about four months into a journey of building out our first exclusively internal-use ambient scribe technology and had roadmapped eventual integration of risk detection and other technologies based on ML and AI.</p><p><strong>Ending 2023.</strong></p><p>The downgrade was concerning. A great many people were relying on us to help them understand and address the impact of neurodiversity on their lives. So we could not allow ourselves to fail. However, the business had weathered major challenges before, and I was confident we had the right skills and determination to meet this one head on.</p><p>Despite the disappointment of the CQC downgrade, 2023 ended on a high note for me with the publication of <a href="https://www.cambridge.org/core/books/digital-mental-health/A987BBB2A6F0350F625A4AF434CC31B3"><em>Digital Mental Health: From Theory to Practice</em></a> in which I had written the chapter on <a href="https://www.cambridge.org/core/books/abs/digital-mental-health/artificial-intelligence/C2589ED3B5993159D72CEDDC05CEDE99"><em>Artificial Intelligence: Why We Need It and Why We Need to Be Cautious</em></a><em>.</em></p><p><strong>All change.</strong></p><p>The new year would see wide-reaching changes and transformation at the business. Our new and dynamic CEO Nell Montgomery built a new executive team. I transitioned from Chief Technology Officer to Chief Artificial Intelligence Officer to focus exclusively on machine learning (ML) and artificial intelligence (AI). This was intended to accelerate implementation of AI and ML, something that was difficult to do as a CTO with broad responsibilities.</p><p>The new leadership team set about ensuring the CQC’s concerns about risks were addressed through a comprehensive risk governance and management programme that would dominate the business through 2024.</p><p>AI would play a pivotal role in this programme, giving us visibility of the risk profile of the thousands of patients communicating with us and boosting clinician productivity while enabling administrative teams to robustly handle referrals.</p><p><strong>An upgrade.</strong></p><p>The CQC made another surprise inspection in February of 2025.</p><p>Although we felt it was too soon, we were in fact ready. The gargantuan effort across the business that brought in a range of partners would result in a rating upgrade from needs improvement to <a href="https://www.cqc.org.uk/location/1-18697798341/reports/LAP-01538/overall">good</a>.</p><p>AI played a defining role in this and I am proud of the team that I have led on this journey. Using results-based management, the output from the team began achieving outcomes and impacts envisioned more than two years before the team’s formation.</p><p>The CQC noted:</p><blockquote>Staff used structured risk assessments and innovative AI tools to monitor patients and prioritise care.</blockquote><p><strong>Return on investment</strong></p><p>Only 5% of organisations investing in AI have seen a <a href="https://db-fastly.legal.io/file/68a930fb07d0d13744418fd2/1755918587.pdf">return on investment</a>. Psychiatry UK is in that 5%. Due to proprietary information, the details are outside the scope of this article. Custom Gen AI apps have been used exclusively within Psychiatry UK to support the patient journey from referral to discharge. The investment in data science has paid for itself many times over in cost savings and other areas while at the same time increasing safety and productivity. This has largely been due to focusing on impact that is aligned with business strategy. To tackle CQC concerns about clinical risks among waiting patients, we deployed AI to surface worrying messages so at-risk patients could receive timely support.</p><p>To reduce the administrative burden on clinicians we implemented ambient scribe before it became fashionable. It generates appointment letters so that clinicians can spend less time on them.</p><p>To manage referrals that were overwhelming staff we developed smart, focused process automations.</p><p>The returns have been in substantial avoided costs, increased patient-facing capacity, improved safety, and better clinician quality of life. Not all return on investment has monetary value, but it should have value. An unexpected benefit was improving job satisfaction — a colleague using our ambient scribe told me:</p><blockquote>I have time and a voice to talk to my family now that I don’t spend so much time dictating letters.</blockquote><p>Though much reduced, I maintain my clinical practice. My belief is that it is important to subject myself to the same technology I subject my colleagues to. So, my colleague’s experience resonates with my own.</p><p>The synergy between trained human clinicians and AI systems in the delivery of care is crucial to the future. It is something I have considered for some time and written about in: <a href="https://www.cambridge.org/core/journals/bjpsych-bulletin/article/clinical-reach-into-the-cognitive-space-critics-outline-conceptual-framework-for-safe-use-of-generative-artificial-intelligence-in-mental-health-decisionmaking/5F7F11857153EFE2D628D08564B41669">Clinical Reach into the Cognitive Space (CRITiCS)</a>.</p><p>This human-AI enablement continues to be at the centre of our approach to AI implementation.</p><p><strong>Another five years of learning.</strong></p><p>A lot has happened in five years. The decision to make this career transition over 10 years ago has given me the privilege of helping not just individuals but thousands of people at a time.</p><p>These are my take-home lessons from the last five years.</p><ul><li>Develop skills in more than one area. It is not necessary to know everything, but you will be more productive if you know more than one thing.</li><li>Choose what you’re optimising for. I optimised for impact. You must understand what you intend to achieve. This is not about winning at all costs but about being clear where you can win and what it will cost.</li><li>Back up your self-belief with knowledge and your knowledge with self-belief. Both are essential.</li><li>Make the most of change. It could unlock your potential and be the opportunity you have prepared for.</li></ul><p>A somewhat surprising experience was that users often expected our AI applications to do more than their intended purpose. This was despite mandatory training being in place for users and access to question-and-answer channels. It highlights the importance of “over-communication”, particularly in a remote work environment and is an area my team and I have recently focused more on. We hope that this will empower our colleagues outside of the core AI team and lead to broader productivity gains.</p><p>Artificial intelligence has come so far in such a short space of time. I believe that in the near future, not using AI in certain healthcare scenarios will be seen as negligent. Much of that AI use will come from focused safe application of generative AI which was in its early infancy just five years ago. There will of course be other paradigms that will shape the future of this technology, its applications and its adoption. The future is unknown but I look forward to what the next year and the next five years will bring.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0f8214381fd0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My journey in the AI XPRIZE]]></title>
            <link>https://lesa-s-wright.medium.com/my-journey-in-the-ai-xprize-dbd5d65bb14e?source=rss-abecb83e6a47------2</link>
            <guid isPermaLink="false">https://medium.com/p/dbd5d65bb14e</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[xprize]]></category>
            <category><![CDATA[personal-development]]></category>
            <dc:creator><![CDATA[Lesa Wright]]></dc:creator>
            <pubDate>Thu, 03 Dec 2020 15:46:04 GMT</pubDate>
            <atom:updated>2020-12-03T15:46:04.122Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Setting a multi-year goal and learning from the experience.</em></p><h4>Planting the seeds</h4><p>In early 2017, I decided to compete in the <a href="https://ai.xprize.org/prizes/artificial-intelligence">IBM Watson AI XPRIZE</a>.</p><p>Six months earlier I had drawn up a five year personal development plan. In the professional dimension of the plan, I had decided to focus on Health Information Technology (Health IT) as the centre piece of a portfolio career in medicine.</p><blockquote>People tend to overestimate what can be done in one year and to underestimate what can be done in five or ten years.</blockquote><p>For some background, I am a consultant psychiatrist (senior doctor) who once worked in the National Health Service (NHS) in the UK. Consultants often take up various “supporting professional activities”. This is work done outside of direct patient contact which is intended to complement and improve patient care and maintain professional development.</p><p>Common activities include teaching medical students and post graduate doctors, medical management, and research. Health IT falls into the medical management arena. In recent years, the Chief Clinical Information Officer (CCIO) role has become more prominent in the UK following the <a href="https://www.gov.uk/government/publications/using-information-technology-to-improve-the-nhs">Wachter Report</a>. Unfortunately, the NHS does not offer real opportunities for clinicians who like me were not only interested in influencing the design, implementation and application of digital health care tools but also wanted hands on development.</p><p>There exists the C<a href="https://www.england.nhs.uk/aac/what-we-do/how-can-the-aac-help-me/clinical-entrepreneur-training-programme/">linical Entrepreneur Training programme</a> in England and the <a href="https://www.bevancommission.org/bevan-exemplars">Bevan Exemplar programme</a> in Wales. This is in addition to various networks around the UK such as the <a href="https://www.ahsnnetwork.com/">Academic Health Science Networks</a>. All these programmes tend to direct innovative clinicians to developers/engineers etc to get their ideas realised.</p><p>Once the decision to focus on Health IT had taken place, I decided to “narrow” it further to artificial intelligence (AI) and machine learning (ML). Why? I had built a rule based mental health records summariser and ML offered the opportunity to rebuild it with fewer false positives and greater flexibility.</p><blockquote>Settling on summarising mental health records to free up clinician time was my chosen project.</blockquote><p>I had been writing software on and off for several years and wanted to get out of the silo in which I worked alone. The decision to focus on Health IT would lead to meeting some interesting people, not just in the XPRIZE but also in the NHS Hackday, Welsh Health Hack, PyData Cardiff and AI Wales.</p><p>The biggest challenge was allocating time to all the skills that needed to be developed. Some general computer science principles, coding, debugging the code, as well as deploying and securing it. All this in addition getting to grips with the legal frameworks of handling patient data and the ethics of automated tools.</p><p>Additionally, being a full-time senior doctor and father meant there had to be a driving force to maintain development momentum of the summarisation tool. That is where the AI XPRIZE came in.</p><p>So I took the risk. There were financial implications, an entry fee, international travel, hotel costs for AI XPRIZE events and of course opportunity costs. Opportunities would be lost for other competitions, missing out on applying to entrepreneur programmes / accelerators and incubators.</p><h4>Competition rounds</h4><p>The first round involved submitting viable project proposals. The initial idea was overly ambitious, going far beyond summarisation. Although the proposal was accepted for its relevance and potential, I quickly realised it needed to be scaled back to make it achievable. Settling on summarising mental health records to free up clinician time was my chosen project.</p><p>The second stage involved demonstrating that the proposal was feasible. The primary outputs of this stage were the performance metrics of the summariser – shown in the figure below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/959/1*gW2X55oAsZy1ARZbwovoAg@2x.jpeg" /></figure><p>The third stage, and also the most intensive, involved a detailed ethical analysis of the potential impact of the proposed artificial intelligence/ machine learning system. It also introduced independent red judges to each competing team. The red judges, experts in their field (such as Natural Language Processing for my project) would help verify that there was no Wizard of Oz and also help teams think more broadly about their solution and its implications.</p><p>Each stage involved writing a detailed academically styled progress report. The reports were scored on pre-defined criteria by independent judges.</p><h4>Fruits</h4><p>My journey in the AI XPRIZE came to an end in the third round at the end of 2019. The field had been whittled down from an initial 600 global applicants to 140 in the first round, 70 in the second, 30 in the third and now 10 semi-finalists.</p><p>The benefit of competing has been the production of a minimum viable product ( MVP ) in addition to a broad professional skill tree. It also meant I got to speak about the project, AI and ML in Cardiff, London, Paris, and Geneva. The summariser became known as <a href="https://www.twynsyght.com/">TwynSyght </a>and the knowledge and skills gained would lead to roles within the Royal College of Psychiatrists, Health Technology Wales, the Medicines and Health Care Products Regulation Authority (MHRA) and now Chief Technology Officer at Psychiatry-UK, the leading provider of online mental health services in the UK.</p><p>I’m pretty satisfied with my outcome. There were competing teams with $Million funding, so I did quite well. The goals of building an MVP and acquiring technical, organisational and administrative skills were met. The transition to portfolio working came two years earlier than hoped — evidence for the adage “People tend to overestimate what can be done in one year and to underestimate what can be done in five or ten years.”</p><p>If I were to do anything differently, it would be to develop a team early and resist the temptation to go it alone. Far greater things can be achieved in groups than in isolation.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dbd5d65bb14e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Speech to text for psychiatrists.]]></title>
            <link>https://medium.com/voice-tech-podcast/speech-to-text-for-psychiatrists-9e311a74f702?source=rss-abecb83e6a47------2</link>
            <guid isPermaLink="false">https://medium.com/p/9e311a74f702</guid>
            <category><![CDATA[alexa]]></category>
            <category><![CDATA[google-home]]></category>
            <category><![CDATA[chatbot-development]]></category>
            <category><![CDATA[conversational-ui]]></category>
            <category><![CDATA[chatbot-design]]></category>
            <dc:creator><![CDATA[Lesa Wright]]></dc:creator>
            <pubDate>Tue, 14 Apr 2020 08:53:15 GMT</pubDate>
            <atom:updated>2020-04-21T11:39:50.013Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sioGfIscmERlcIsaOeqV5Q.jpeg" /><figcaption>What are the limits of speech to text transcription in healthcare?</figcaption></figure><h3>Speech to text for psychiatrists and other healthcare professionals.</h3><p><em>I work as a consultant psychiatrist and would like to share my experience with speech to text transcription to improve my productivity in clinical work.</em></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fbingewith.com%2Farticle%2F%3A%3A_Generate_Audio&amp;dntp=1&amp;display_name=BingeWith&amp;url=https%3A%2F%2Fbingewith.com%2Farticle%2F%3A%3A_Generate_Audio&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=bingewith" width="480" height="110" frameborder="0" scrolling="no"><a href="https://medium.com/media/d6a9e202d4ee773dbc23199fcf6d186c/href">https://medium.com/media/d6a9e202d4ee773dbc23199fcf6d186c/href</a></iframe><p>About 15 months ago, needed to reduce the turnaround time for clinic letters. Up until that point I would dictate a letter on an analog dictaphone and this would then go into a typing queue. It could often take up to 2 months for a letter to be typed because of the pressure my secretary was under to perform various service -related admin duties. I would then suddenly have a flood of letters to check and little time to review them in.</p><h4><strong>Google Speech to text transcription.</strong></h4><p>I decided to give speech to text services ago. My first option was to go with the Google transcriber + google docs on my phone. This made a lot of sense as it is ”free“, and I could get started straight away.</p><p>The first issue that should come to mind here is the privacy of patient data. I got around this by using pseudonyms for my patients. I used the same handful of pseudonyms for all my clinics. As a result I now have records of Bruce Wayne being treated for bipolar disorder, schizophrenia and personality disorder amongst other conditions. I also made a point not to mention the names of any colleagues, family members or other people involved in an individual’s care. Lastly, I also did not refer to geographical locations and use pseudonyms for common locales.</p><blockquote>The macro worked with a single click, automatically correcting these errors throughout the text which saved a lot of time.</blockquote><p>At the end of a clinic I would send all the dictations to my work email, copy these into Word and use the find and replace feature to replace the pseudonyms with the correct names. This process was quite tedious so I put together a Word macro — this is a program that you can write inside of Word and other office products that you can use to add new features. This corrected common transcription errors. For example Google’s speech to text often made mistakes with the phrase “new paragraph” by transcribing it instead of moving the cursor to a new paragraph. It did the same with the phrases open quote and close quote rather than inserting quotation marks.</p><p>The macro worked with a single click, automatically correcting these errors throughout the text which saved a lot of time. However, this was only suitable for some errors. For example Google often replaced “he” with “it” and “her” with “the”. These couldn’t simply be replaced as there would frequently be false positives.</p><p>On the whole it worked quite well and instead of having to wait a few weeks for letters to be ready the main body of a letter was generally ready the same day or within a few days depending on other pressures.</p><blockquote>Build better voice apps. Get more articles &amp; interviews from voice technology experts at voicetechpodcast.com</blockquote><h4><strong>QuickStart to using a Word macro.</strong></h4><p>Here is some guidance to get you going — an explanation of the macro is beyond the scope of this article and I am by no means an expert in Word macros:</p><p>· Open Word&gt;View&gt;Macros &gt; View Macros</p><p>· Enter a new name (I used DictationFix) in the first field of the popup window, then click create. A macro editor will open.</p><p>· Download the dictation fix text file from <a href="https://bitbucket.org/lswright/dictationfix/downloads/">here</a> (click download repository, and open the zip file to find the dictationfix.txt file). A note on security, if you don’t know what a macro does, don’t run it inside Word. Having said that, this macro is simple and there are links to explanations of what the code means in the text file.</p><p>· Open the text file, select all (ctrl+a), copy (ctrl+c) the macro script.</p><p>· In the macro editor paste (ctrl+v) the script</p><p>· Click save</p><p>· Click the run key to execute the macro on your text.</p><p>To run the macro later:</p><p>· Open Word&gt;View&gt;Macros &gt; View Macros</p><p>· Select DictationFix and click run</p><h4><strong>Nuance Dragon Dictation Anywhere.</strong></h4><p>After using Google for about eight months, I switched to Nuance Dragon dictation anywhere. This has a relatively low monthly subscription charge and overall I have found this to be a better experience than using Google dictate. I have not used the replacement script as the kinds of errors I mentioned earlier do not occur with it. It also struggles with “he” and “it” sometimes but performs better than Google did.</p><blockquote>Medications likely to appear in popular media transcribed correctly. Most symptoms also transcribed correctly.</blockquote><p>Additionally being able to add vocabulary has helped reduce the error rate. You can also use single phrases to add whole blocks of predefined text. The Dragon application transcribed most medication names correctly (antidepressants, antipsychotics, anticonvulsants and benzodiazepines) particularly if generic names rather than brand names were used. Medications likely to appear in popular media transcribed correctly. Most symptoms also transcribed correctly. As example additions, I had to add the word “euthymic” to the vocabulary and branded names of medication such as Concerta XL.</p><h4><strong>Microsoft Word dictate.</strong></h4><p>The performance of Microsoft Word dictate is about the same as Google dictate. I have found it useful for making quick corrections or additions (of more than a few words) to letters as I can still speak faster than I can type. However it performs very poorly with technical terms and there does not appear to be a way to modify the vocabulary.</p><h4><strong>Practical considerations.</strong></h4><p>It is important to speak clearly at a moderately steady pace and it helps to emphasise words. Dragon Dictate works at arm’s length which is handy as I can just put the phone on my desk and read my notes. It is more efficient not to repeatedly check every single phrase/sentence after dictation. If you’re dictating from notes it is useful to simply finish dictating otherwise the dictation takes much longer than it should. Leave corrections to the end. The less ambient noise there is, the better the performance.</p><p>Microsoft Word’s transcription works better with a decent quality microphone.</p><p>In terms of measurable productivity, I found that it takes me between 3 1/2 to 5 minutes to dictate a single page of my handwritten notes. So my Goldilocks zone is really 1 to 2 pages of notes as might occur in an outpatient review for speech to text transcription and anything longer (for example an initial assessment or a court report) than resulted in time lost with checking and therefore I prefer to send that to a human transcriber.</p><p>Overall, I have found speech to text dictation to be a good productivity boost but needs proofreading and it is helpful to use an auto correction macro in word (or equivalent) if required. The nuances of pronunciation errors are likely to differ from person to person so ultimately experimentation is required.</p><h3>Something just for you</h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Flinks.collect.chat%2F5d8c870c8705ce12e45fb7d0&amp;dntp=1&amp;display_name=Collect.chat&amp;url=https%3A%2F%2Flinks.collect.chat%2F5d8c870c8705ce12e45fb7d0&amp;image=https%3A%2F%2Fcollect.chat%2Fassets%2Fimages%2Flogo-green.png&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=collect" width="600" height="600" frameborder="0" scrolling="no"><a href="https://medium.com/media/7f5cb45fd785102f7af729c560e2de3f/href">https://medium.com/media/7f5cb45fd785102f7af729c560e2de3f/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2F6b7b35%3Fas_embed%3Dtrue&amp;dntp=1&amp;display_name=Upscribe&amp;url=https%3A%2F%2Fupscri.be%2F6b7b35%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/8a76a1650cf30355bd307195d382558a/href">https://medium.com/media/8a76a1650cf30355bd307195d382558a/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9e311a74f702" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-podcast/speech-to-text-for-psychiatrists-9e311a74f702">Speech to text for psychiatrists.</a> was originally published in <a href="https://medium.com/voice-tech-podcast">Voice Tech Podcast</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>