AI in Speech Therapy: What 300 Hospitals Taught Us (and What You Can Use Today)
You’ve probably already used AI this week. Maybe you asked ChatGPT to help you word something, or your meeting software summarised your notes automatically. So it’s not a big leap to wonder: what’s AI actually doing in speech therapy right now?
In this episode of Bilingual Speech Spot, Chloe sat down with Linda Chow — a bilingual SLP and Clinical Development Lead at Peisheng International — to talk through what’s already in use, what’s coming, and where the limits are.
What We Talked About AI in Speech Therapy
Linda shared insights from the ASHA conference in Washington DC, where AI-assisted language sampling and diagnostic tools were a major theme. She also pulled from her own clinical work across the US and China — including Peisheng’s decade-long project scaling speech therapy services through a research-first approach.
The conversation ranged from the practical (how SLPs are using ChatGPT to generate therapy materials) to the cutting-edge (AI video analysis that counts a child’s eye contact in real time).
Three Things That Stood Out
1. The tools SLPs are already using
If you’re a speech pathologist, chances are you’re already using some form of AI — even if you don’t call it that. Linda described how clinicians are using general tools like ChatGPT to generate word lists tailored to a child’s phonological targets, and clinical tools like D-Health to produce a session note backbone that the therapist then refines. Fathom AI is being used to summarise meetings. These aren’t futuristic — they’re Tuesday-morning workflows.
The key message: use AI for tasks where errors are catchable. A clinician reviewing an AI-generated word list will spot a poor fit immediately. A parent who doesn’t have clinical training won’t always know when the AI output is off.
2. What Peisheng’s model actually shows us
Peisheng’s work is a useful case study in how to scale speech therapy services without cutting corners on quality. Starting with the DreamSea screener — the first standardised Mandarin language assessment, published in 2015 — the model expanded to 300+ hospitals by prioritising research first, clinical buy-in second, and scale third.
New tools coming from this approach include MENT AI, an automated Mandarin narrative assessment tool that uses voice recognition to separate child and therapist speech, and SWAM AI, which analyses phonological patterns like stopping and produces instant reports. These tools don’t replace SLP judgement — they compress the time it takes to get from assessment to clinical interpretation.
3. Phenotyping — the one that surprised us
Linda introduced a technology called phenotyping: AI analysis of observable behaviours from video recordings. It can track eye contact, gesture use, pointing frequency, and communication exchanges — frame by frame, at 30 frames per second. This has obvious implications for autism assessment, parent coaching, and classroom observation. It’s also a reminder that AI can catch patterns humans miss, precisely because it doesn’t get tired or distracted.
What Practitioners and Parents Can Do
If you’re an SLP:
- Explore what AI tools are already available in your documentation system — many are built in
- Use ChatGPT for materials generation, not clinical decision-making
- Stay informed about tools like MENT AI and SWAM AI as they become available in English and bilingual contexts
- Keep professional judgement at the centre — AI is a co-pilot, not a replacement
If you’re a parent:
- AI can be a useful starting point for understanding speech and language development — but treat it like a search engine, not a diagnosis
- Large Language Models (like ChatGPT) are pattern-matching tools, not clinical tools. They don’t observe your child, they can’t detect what’s missing from an interaction, and they sometimes get things wrong in ways that aren’t obvious
- Use AI to help you form questions to bring to your SLP — not to skip the appointment
The most important point Linda made: AI extends reach. Communities where access to speech therapy is limited — whether because of geography, funding, or workforce shortages — stand to benefit most from well-designed AI tools. That’s the version of this story worth watching.
To hear the full conversation, listen to Bilingual Speech Spot Episode 18 on Spotify or wherever you get your podcasts.
Visit blackburnslp.com.au/resources | Follow @phd.speechie.mum on Instagram
AI寫報告、做評估——言語治療師應該驚定應該用？從300間醫院學到的事
如果你是言語治療師，大概已經搜尋過「AI可以幫言語治療師做什麼」，或者在同事群組裡討論過ChatGPT會否取代我們的工作。這一集《Bilingual Speech Spot Podcast》第18集，Chloe與Linda Chow（雙語言語治療師、裴昇國際海外臨床部主管）對談，梳理AI工具現時在言語治療診所裡究竟在做什麼，以及未來會帶來什麼改變。
治療師現在已經在用的工具
如果你是言語治療師，你可能已經在使用某種AI——即使你沒有稱它為AI。Linda提到，臨床人員用ChatGPT為特定孩子生成詞語清單，用D-Health整理治療記錄的基本框架再自行修改，用Fathom AI整理會議摘要。這些不是未來的事——是每星期的日常工作。
重點在於：AI最適合用在錯誤可以被發現的工作上。一位受過臨床訓練的治療師看到AI生成的詞語清單，很快就能判斷是否合適。但沒有臨床訓練的家長，未必能識別AI的輸出何時出錯。
裴昇模式教會我們什麼
裴昇的工作是一個很好的案例——如何在不降低質素的情況下大規模擴展言語治療服務。從2015年發布首套普通話標準化語言評估工具DreamSea開始，以科研先行、臨床支持、規模其後的方式，最終擴展到中國300多間醫院。
新工具包括MENT AI——一套自動化普通話敘事評估工具，透過語音識別技術區分兒童與治療師的說話；以及SWAM AI，可以即時分析語音問題，直接生成報告。這些工具並非取代治療師的判斷，而是壓縮由評估到臨床解讀的時間。
Phenotyping——最令人驚訝的技術
Linda介紹了一項稱為phenotyping的技術：以AI分析影片中的可觀察行為。系統可以逐格（每秒30幀）追蹤眼神接觸、手勢使用、指向動作與溝通交換次數。這在自閉症評估、家長訓練與課室觀察方面都有明顯的應用價值，同時提醒我們：AI能發現人類容易錯過的規律，因為它不會疲倦，也不會分心。
給治療師的建議
- 了解你現時使用的系統中有沒有內置AI功能
- 用ChatGPT生成訓練材料，不要用它做臨床決策
- 持續留意MENT AI與SWAM AI在英文及雙語版本的發展
- 保持臨床判斷在中心——AI是副駕駛，不是司機
- 如果你想與其他治療師交流AI在臨床應用上的實際經驗，歡迎加入 Bilingual Spot Special Interest Group，延續這一集的討論。（連結：加入表單）
給家長的話
如果你也想知道AI會否影響孩子的言語治療評估——Linda的建議是：AI可以是了解語言發展的起點，但不要把它當成診斷工具。大型語言模型是模式配對工具，不是臨床工具，它不能觀察你的孩子，也未必能發現互動中缺少了什麼。用AI幫助你整理問題，再帶去見你的言語治療師，而不是取代預約。
Linda最後強調的一點最重要：AI能夠擴展可及性。對於因地理位置、資金或人手不足而難以獲得言語治療服務的社區來說，設計良好的AI工具最有意義。
完整訪談請在Spotify或你常用的Podcast平台收聽《Bilingual Speech Spot Podcast》第18集。
瀏覽 blackburnslp.com.au/resources ｜ Instagram：@phd.speechie.mum