---
id: "workshop:reading-out-loud-growing-strong-ai-tools-for-fluency-development"
type: "InnovateUS workshop"
title: "Reading Out Loud, Growing Strong: AI Tools for Fluency Development"
url: "https://innovate-us.org/reading-out-loud-growing-strong-ai-tools-for-fluency-development"
date: "2025-03-12"
duration: "60 minutes"
language: "English"
series: "AI for Literacy Achievement"
instructors: ["L. Burleigh"]
recording: "https://www.youtube.com/watch?v=F3zs3PGulg8"
transcript: "automatic, from the recording"
source: "InnovateUS (innovate-us.org)"
licence: "CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org"
---

# Reading Out Loud, Growing Strong: AI Tools for Fluency Development

InnovateUS workshop · 12 March 2025 · 60 minutes

Focus on mastering reading fluence using AI-powered tools that leverage Automatic Speech Recognition (ASR) to guide read-aloud practice and provide personalized feedback. Fluency is a critical component of literacy, and ASR technology is enabling students to practice reading aloud with immediate, tailored guidance. Through hands-on experience, participants will explore ASR-driven tools for literacy instruction to support diverse learners, improve reading fluency, and foster a love of reading.

## Learning goals

- Focus on mastering reading fluence using AI-powered tools that leverage Automatic Speech Recognition (ASR) to guide read-aloud practice and provide personalized feedback
- Explore ASR-driven tools for literacy instruction to support diverse learners, improve reading fluency, and foster a love of reading

## Speakers

- **L. Burleigh**, Data Analyst, The Learning Agency

Part of the series [AI for Literacy Achievement](https://innovate-us.org/workshop-series/ai-for-literacy-achievement).

Recording: https://www.youtube.com/watch?v=F3zs3PGulg8

## Transcript

Automatic transcript of the recording: names and terms can be misheard, and the times can be a few seconds off. Each time links to that moment of the video. The transcript alone: https://innovate-us.org/reading-out-loud-growing-strong-ai-tools-for-fluency-development.txt (plain text) and https://innovate-us.org/reading-out-loud-growing-strong-ai-tools-for-fluency-development.json (every segment with its start and end second).

**David Fields**

[0:00](https://youtu.be/F3zs3PGulg8?t=0) Thank you all for joining us for the fourth workshop of Rethinking Reading. This is the AI for Literacy Achievement Workshop series and this is an Innovate US series sponsored by the Burns Center for Social Change's Equitable Literacy Initiative and led by our partners at the Learning Agency. So in this series we are showcasing how innovative leaders are using AI to make rapid gains in advancing literacy in today's classrooms and you can find the full schedule and register for all the upcoming workshops at the Innovate US website. We'll put the link in the chat so that you have that. My name is David Fields. I'm a senior fellow at the Burns Center for Social Change at Northeastern University and the Burns Center designs and implements practical solutions to society's hardest problems. We develop innovative and participatory and equitable approaches to solving public problems using innovative new technologies. So for those of you not yet familiar with Innovate US, a program out of the Burns Center, I encourage you to visit innovateus.org and check out. There's hundreds of free 60-minute workshops and at-your-own-pace courses in AI and digital skills designed for public sector professionals. So now I'm thrilled to hand it over to Elberle, a data analyst at the Learning Agency, where they bring a background in statistical data analysis and human behavior research focusing on cognition and learning. Before joining the Learning Agency, they performed research and data analysis in cognitive neuroscience as a graduate quantitative researcher and investigated analytic assumptions in machine learning and transfer learning algorithms, fear learning, and generalization of knowledge. They are also experienced in instructional design and have created and taught courses and labs on statistical data analysis in R. El holds a PhD in

[2:01](https://youtu.be/F3zs3PGulg8?t=121) cognitive and brain science from Louisiana State University and a BS in psychology from the University of New Orleans. Without further ado, El, I'm going to turn it to you. Thank you.

**L. Burleigh**

[2:13](https://youtu.be/F3zs3PGulg8?t=133) Thank you so much for that introduction, David. I'm going to go ahead and share my screen and we can get started. Okay, so welcome to today's workshop. We're talking about mastering reading fluency through ASR-guided read-aloud practice and personalized feedback. So, today's workshop is one of many in a series that's aimed at helping education innovators explore how AI can advance K-12 literacy. We're looking to provide skills with implementing AI literacy tools, create personalized reading intervention strategies, and design inclusive literacy programs. And like David mentioned previously, I'm from the Learning Agency, which is hosting these workshops in partnership with InnovateUS and the Byrne Center for Social Change's Equitable Literacy Initiative.

[3:18](https://youtu.be/F3zs3PGulg8?t=198) So, today we're going to explore how ASR or automatic speech recognition tools can enhance students' reading and literacy skills. So, ASR tools can provide assessments and personalized feedback to support reading fluency and guide students' literacy understanding. We're going to discuss how various AI tools can help readers build stronger reading skills and foster better learning outcomes through immediate and interactive methods. So, these ASR methods transform spoken word into text in real time. That's what allows AI to listen attentively as children read aloud and provide that instant feedback to guide their progress. So, our objectives today include an expert deep dive on history and trends in ASR. We're going to explore some future directions, key research questions, and the Learning Agency's approach within the space. We'll talk a little bit deeper about two popular AI gamified tools, sorry, two popular ASR tools that do also have gamified techniques for targeted feedback and dynamic learning. And we're going to explore some additional resources and strengths of one of the tools doing a little activity. So, starting with the history or evolution of ASR, the first system was developed in 1952 when Bell Laboratories designed a system to recognize a single voice speaking digits through a voice track resonance. So, that was our very early ASR systems, but it wasn't until the 1980s where we really had a pivotal decade where we got a statistical method called the Hidden Markov Model. And this model was introduced and revolutionized language

[5:19](https://youtu.be/F3zs3PGulg8?t=319) modeling, improved accuracy, and laid the groundwork for more sophisticated ASR. This model really enabled prediction of the most probable phonemes to follow a given phoneme. It allowed for that prediction and for that track to be assessed. During this time, IBM introduced a voice-activated typewriter with a 20,000 spoken word vocabulary, and the first fully interactive toy named Julie was released that responded to and generated basic speech. In the 1990s, the emergence of microprocessors enabled faster and more accurate speech pattern recognition. Dragon Dictate became the first speech recognition software specifically designed for consumer use, which was released in 1990. And in 1997, Dragon then developed the first continuous speech recognition product capable of understanding up to 100 words per minute of continuous speech. So before this, systems really relied on discrete dictation or pausing after every single word to ensure accurate recognition, which isn't how we speak in the real world.

[6:45](https://youtu.be/F3zs3PGulg8?t=405) In 2007, Google first attempted speech recognition with a telephone-based directory service, and that data really helped Google improve the dictation systems that we know and use today. A few other statistical methods also emerged during the early 2000s that further improved accuracy, such as deep learning methods that really decreased word error rate and the ratio of errors in a transcript to the total words spoken to measure speech recognition accuracy. That's our word error rate. It's decreased by 30% with some of these methods. There's a large jump. So pre-2000s and the first handful of the years of 2000s were a bit slow with difficulties really being in the lack of big training data and big computing power until that deep learning surge right around 2009-2010. So then once we really get into the 2010s, we had a really big rapid rise in smartphone market penetration, which really progressed our technology.

[8:03](https://youtu.be/F3zs3PGulg8?t=483) So during this time, by the time 2020 came around, 72.2% of the population had a smartphone, which really increased the speech recognition software and apps. It also sparked smart speakers and digital assistants like Siri, which launched in 2011. During this time, Google's algorithms achieved a 95% English word accuracy rate, which was equivalent to humans. And really during this time, the distinction between speech recognition and sorry, speech or voice recognition and speaker recognition came about, which was a major breakthrough to be able to distinguish between different voices. And as we're getting into the 2020s, we have a big influence of AI on ASR technology that's really transforming the space. So optimization techniques have reduced training and prediction time, which are pretty high. And generative AI are increasing human digital interaction and models for audio processing, voice cloning, and context-aware applications. Conversational AIs with personal assistants like Alexa and Siri are also growing and supporting more nuanced communication and multitask learning frameworks for a wide array of spoken language tasks.

[9:34](https://youtu.be/F3zs3PGulg8?t=574) And automated captions have also increased accessibility during this time. So there's some leading AI developers and ASR tools that are advancing technology by using the latest technologies, techniques, and AI algorithms for maximum optimization and to support literacy learning with different learning needs, such as in dyslexia and English as a second language. For example, Soapbox from Scholastic is built on tens of thousands of hours of children's speech from 193 countries. These sort of platforms aim to incorporate innovative features to further reach and support each literacy learning journey, such as voice-enabled assessments, to optimize student grouping and enhance the quality of small group and individualized instruction, reports on aspects such as usage, performance, and proficiency levels, real-time output to deliver immediate feedback, and structured dialogue for high interaction and fluency improvement. And we're going to dig into some of these a little more as we get more into our tools. But these platforms really integrate AI to deliver personalized feedback and provide immediate constructive feedback as readers are reading aloud and as they struggle or mispronounce a word to really work on reinforcing learned reading skills, vocabulary, fluency, and self-correction. So these ASR platforms act as a personal intelligent reading tutor through large libraries of text for increased relevance and interest to each reader and provide an active role in the reader's literacy learning.

[11:37](https://youtu.be/F3zs3PGulg8?t=697) Some key research questions and directives include ways to make tools more inclusive and accurate to perform highly on more dialects, accents, and languages, while ensuring data privacy with the student's speech data, and assessing technology considerations. So we've progressed technology really far and fast in recent years, but there are still limitations that come about within this ASR technology. For example, looking at linguistic differences of full language, accent, and dialect coverage is a significant challenge for speech recognition software, given that there's over 7,000 languages spoken in the world, each with many accents and dialects. English, for example, has more than 160 dialects around the world. In a study on the voice technology barriers in 2020, actually found that 66% of the respondents found accent or dialect related issues when they were adopting voice recognition technology.

[12:48](https://youtu.be/F3zs3PGulg8?t=768) Another key consideration is inclusivity. So increasing accessibility for very linguistic differences is an inclusivity research focus, as is the efficacy of speech recognition for children with disabilities, such as various learning disabilities and vocal or speech disabilities. So technology is moving forward and addressing many of the problems, such as speech recognition software now often including voice profiles to become more accurate at distinguishing words spoken particularly by younger users, which has been a high limitation for ASR tools. Younger speakers are more difficult for these machines to pick up. But given the accuracy on children's speech is recent because we more so started training on children's speech, we do need more research within these younger populations to really look at how ASR is working.

[13:58](https://youtu.be/F3zs3PGulg8?t=838) Privacy and security are a big question, especially whenever we get into children's privacy, children's data here. Because a voice recording is a person's biometric data, which is sensitive and has privacy related issues attached. Again, especially for children. So ASR tools need a large amount of varied data, which means a large amount of sensitive children's biometric data. So given the high security and privacy of both it being children data and biometric data, the data that's needed to train and produce ASR tools is difficult to access and requires high consideration of the data's storage and use. The model's optimization is also always a consideration and a direction. So achieving a high accuracy of speech recognition is critical.

[15:00](https://youtu.be/F3zs3PGulg8?t=900) Speech recognition is very challenging. Not only do the wide variety of linguistic differences and limited data access pose challenges in increasing model accuracy, but other speech and auditory aspects also impact performance and optimization of these algorithms. For example, background noise such as crosstalk and white noise that appears in the real world can disrupt the speech recognition software. And things like field and regional specific jargon can be difficult for a model to understand as well. Looking at cost and hardware, developing ASR technology can be costly and is an ongoing process. So, data collection and storage can be expensive, time-consuming, and take a large amount of space. Developing the model requires strong computational power, which requires the appropriate hardware. And deployment of the tool is also difficult and expensive to meet those device requirements.

[16:06](https://youtu.be/F3zs3PGulg8?t=966) And even once they're available, time and resources are continuously needed to improve the accuracy of the model and expand it to cover additional languages and other linguistic differences as these technologies advance. And finally, looking at human-AI collaboration, studies are also exploring how AI can assist in literacy tasks and learning to provide support and aid teachers, but maintaining that human oversight. So, we're really getting into what's best for both children and teachers with using these tools. So, I have a statistic down here too. In 2019, before the pandemic, the U.S. National Center for Education Statistics, or NCES, reported that only 35% of fourth graders performed at or above their reading proficiency level.

[17:07](https://youtu.be/F3zs3PGulg8?t=1027) However, if we take a look a little deeper, it would take two-thirds of a teacher's working day to spend only 15 minutes of one-on-one reading practice with each student in a class of 25. So, this is really something that time is super helpful here with these tools. The Learning Agency is exploring the ASR space by publishing on our extensive ASR research reviews and individual interviews with ASR research experts regarding children's ASR data, the considerations, methods, and modeling needs to really increase awareness of the current challenges and limitations of the field. Our Learning Engineering Tools Competition also has a new track this year called the Dataset Prize Track, which is to support the open-access release of innovative education datasets and has the potential to fund a new ASR dataset.

[18:10](https://youtu.be/F3zs3PGulg8?t=1090) And finally, we have a lot of extensive experience with education-based data science competitions, and we are providing support to the development of an upcoming children's ASR data science competition to facilitate the increase of some of these algorithms and what these tools can do. A few big tech companies, such as Google and Microsoft, are developing speech tools to engage young readers in literacy education through large selections of engaging stories to read, real-time feedback as the tools listen to children reading aloud, and celebrating achievement through gamified features like points. Read Along is by Google and uses advanced text-to-speech and voice recognition technologies to listen and respond to real-time feedback.

[19:11](https://youtu.be/F3zs3PGulg8?t=1151) It has thousands of illustrated stories in 11 different languages, gives prizes and badges for motivation, and allows groups to be created by teachers or parents to track the activity and progress of students. Learning Coach is by Microsoft and provides targeted practice by detecting challenging words and automatically incorporating additional opportunities to practice them. It also allows readers to co-create their own stories using AI to increase engagement and relevance to the readers. And this tool encourages progress through methods such as unlocking new story elements as the learners read.

[20:12](https://youtu.be/F3zs3PGulg8?t=1212) So a quick flag I'd like to have here is that though these two tools may not be as extensive as some of the other education ASR tools that we're going to look at, I would like to note that these are free tools, whereas some of the others aren't. Because of those limitations we discussed earlier with a high cost, the high security, some of these tools are less financially accessible, but these two tools are. So the first tool that we're going to dig into is Amira Learning. Amira Learning is an intelligent reading assistant built on cognitive science and the 20 key science of reading principles. It uses an assess, instruct, tutor or AIT method, which starts by assessing students with a 15 to 20 minute assessment to use as a benchmark test, a universal screener that's dyslexia screener and to progress monitor and to monitor progress of the readers.

[21:23](https://youtu.be/F3zs3PGulg8?t=1283) So instruction is provided with an AI assistant with the intention to simplify lesson planning and assessment management for teachers. And the tutoring provides science of reading based coaching while students read aloud, keeping them in their zone of proximal development. Amira Learning does have multilingual growth. So both bilingual and Spanish literacy programs are available in Amira Learning. Bilingual programs incorporate cognates, which are words with shared meanings across languages, into the adaptive instruction to use their native language to increase English proficiency. The Spanish literacy methods increases reading and fluency, increases reading fluency and literacy for native Spanish speakers by establishing a strong foundation in the student's native language.

[22:26](https://youtu.be/F3zs3PGulg8?t=1346) So Amira also provides teachers with an intelligent assistant to assess skill gaps, provide high quality evidence based instructional materials, it optimized student groupings and will act as an AI tutor, providing tailored micro interventions and micro lessons to the students. So Amira can continuously monitor student progress without requiring a teacher to sacrifice their instructional time. And differentiated assignments can be seamlessly integrated into the teacher's core curriculum. So Amira has been researched by universities such as Carnegie Mellon University and Columbia University and regularly results in high effect sizes, demonstrating high impacts of its literacy growth approach. The next tool we're going to dig into is LUCA AI, which is also a digital reading tutor with an immersive reading platform, designed based on research by experienced researchers and renowned academics, such as the Yale Center for Dyslexia and Creativity, which aims to advance research and understanding of individuals with dyslexia, really focusing on research on identifying, understanding and instructing people with dyslexia.

[23:51](https://youtu.be/F3zs3PGulg8?t=1431) They also work on developing interventions and providing resources for individuals with dyslexia. Another big research point is the Dyslexia Library, which is part of the International Dyslexia Association, provides information on dyslexia and other learning disabilities through an online database of articles, books and other research materials, providing resources and tools for parents, teachers and individuals with dyslexia. And additionally, there's some research by Texas A&M, they based off of on the prevalence of reading fluency and vocabulary difficulties in adolescents that struggle with reading comprehension, which examined the impact of morphological awareness on reading abilities in individuals with dyslexia. This study found that the morphological awareness should be taken into considerations when designing interventions for individuals with dyslexia, as they found it plays an important role in improving their reading ability.

[24:58](https://youtu.be/F3zs3PGulg8?t=1498) So there is some more research that LUCA AI is based off of, and they have all that information on their website, if you'd like more details. So LUCA AI uses AI in multiple ways to deliver engaging and personalized literacy learning. One method is through the StoryLab program, which can generate personalized stories tailored to the reader's interest, which incorporates both the reader's level and the challenges that they're having to encourage engagement with a personalized learning experience to increase reading retention. LUCA Listens is innovative technology that uses advanced speech recognition to identify the letter-sound pairs that are challenging for a learner, and to customize their learning experience with real-time interventions for improved reading comprehension. And Professor AI provides personalized lesson plans and AI-powered reading recommendations that are tailored to each student's individual needs to ensure they are reading the right material to reach their goals and help them learn faster.

[26:17](https://youtu.be/F3zs3PGulg8?t=1577) So we're now going to look a little deeper at the Amira platform. I'm going to stop sharing and re-share another screen real quick. As mentioned, the Amira platform is not a free tool, so we're not going to look at, we're not going to dig into an account itself like we did previously, if you've come to others before, but we're going to look a little bit about some of the materials they have looking a little deeper. So looking at the process of Amira, and these tools can vary in this process, but many of them follow the same general guides. For the student reads aloud and the algorithm analyzes their performance in real time, Amira specifically uses AI detects in reading pass. So if you're not familiar, RAN is a rapid automatized naming task.

[27:22](https://youtu.be/F3zs3PGulg8?t=1642) It's about speed of speech, and it uses nonsense word fluency tasks to really help detect where a student is having trouble. It will also flag at-risk students and provide instant insights, particularly within the teacher view, it'll create great reports to really look at where students are struggling. Also within the teacher side, it will generate instructional guidance that's aligned with the scope and sequence of the teacher. And really one of the big parts is delivering in the moment tutoring. So while the student is reading, and while it's listening to the reading aloud, the tool will use micro-interventions to identify and help when a student is struggling. So we're going to watch a short video, a little demo of the tool, and so you can see how it interacts with the student.

[28:39](https://youtu.be/F3zs3PGulg8?t=1719) So we're going to watch a short video, a little demo of the tool, and so you can see how it interacts with the student.

**David Fields**

[29:24](https://youtu.be/F3zs3PGulg8?t=1764) I'll just double check, and we're not hearing anything. We see the closed captions. Are we supposed to be hearing anything?

**L. Burleigh**

[29:30](https://youtu.be/F3zs3PGulg8?t=1770) Yes, I was hoping so. Let's see. My apologies. Do you hear anything now, David?

**David Fields**

[29:54](https://youtu.be/F3zs3PGulg8?t=1794) No, no audio.

**L. Burleigh**

[29:56](https://youtu.be/F3zs3PGulg8?t=1796) Okay.

**David Fields**

[30:00](https://youtu.be/F3zs3PGulg8?t=1800) Here you go. Super. Let's keep going. He will go away. He is soft. Funny is a pretty hard word. Which picture best shows the word funny? Let's go to work on stay. Stay is an important word. Say stay three times as fast as you can.

[31:01](https://youtu.be/F3zs3PGulg8?t=1861) Stay, stay, stay, stay, stay. Got it. Just stay. I'm going to sound this word out. Near makes near. Near. I like. And I'm going to move forward and you'll see here that Amir is intelligent enough to bring a word back that the student struggled with earlier to see if they're able to master it later in the reading session. But just don't, do not, you, touch. The students master it the second time around.

**L. Burleigh**

[31:58](https://youtu.be/F3zs3PGulg8?t=1918) Okay, so as we can see, whenever Amir listens to the student speaking, and it can identify when the student is having trouble and jump in with some of these micro interventions that are engaging for the student to help them learn the material better and really help it solidify. So the way Amir works is through working on the reading rope, which is part of the science of reading. If you're not familiar, the reading rope has two primary sections, it's based on word recognition and language comprehension. So they work on phonemic awareness, really working on the sound phonics, so really working on building the ability to put these sounds together. Fluency is another big point, all about the rhythm and flow, working on putting all of these words together to make fluent statements, fluent reading.

[33:09](https://youtu.be/F3zs3PGulg8?t=1989) Vocabulary, helping to expand the vocabulary as students read and make connections between various words that may have various meanings of the same word. And comprehension, putting all of this together and really allowing students, helping students put all the information together to have a deeper understanding of what they're reading. And finally, I'm just going to show a little bit, I thought the teacher view is really interesting here. So I wanted to point it out. Let's see if we can, oh, the image doesn't actually get bigger, ah, there we go. So this is one of the teacher views. So you can see with weekly instruction, Amir does work with both class and it can be integrated at a district level. It looks at the various curriculums that you're using and really helps teachers align on what their students need.

[34:18](https://youtu.be/F3zs3PGulg8?t=2058) Particularly in this view, we can see where the assignments are broken down. And notice the AI here is jumping in to assess how students are doing with the program across the students to help identify where additional instruction may be helpful. So you can see that the AI says that three students could use more instruction on knowing the primary letter sound correspondence for the consonant, consonant M. So this program can really help dive into what is still needed for students to understand where they're struggling to guide instruction rather than replace instruction. And we're going to go back to wrap up our slides.

[35:23](https://youtu.be/F3zs3PGulg8?t=2123) Share my screen here again. So our primary takeaways here, we talked about evolution of ASR tools, which have progressed from very limited discrete dictations, meeting all those pauses, to advanced generative models that can dictate the continuous speech and has much higher accuracy than previous models. We talked about educational data science for innovation with ASR data science competitions being supported and the compilation of ASR methodology and data. And we talked about innovative tools like Read Along, LUCA AI, and others in order to provide personalized feedback and content aligned with literacy learning research and create inclusive and supportive learning environments.

[36:23](https://youtu.be/F3zs3PGulg8?t=2183) And we looked at how ASR tools can help educators by providing real-time data and research-backed resources to aid students with their learning journey. And as we mentioned in the beginning, I am Elle Burley with the Learning Agency, and the next workshop in this series is the AI-powered screening tools for dyslexia, which will be on March 25th. We do have a newsletter that goes out that's all about the latest in edtech, so feel free to check that out to stay up to date with what's going on there. Thank you.

**David Fields**

[37:04](https://youtu.be/F3zs3PGulg8?t=2224) Wonderful. Thank you, Elle. That was fantastic. So we're going to turn to questions here, because some came in while you were presenting, and I want to make sure that we've got an opportunity to kind of walk through these a little bit. So going back up, Leslie, feel free to take yourself off mute if you want, but you had a question about bands.

**Audience**

[37:30](https://youtu.be/F3zs3PGulg8?t=2250) Yes, hi. I'm primarily secondary, so I'm just wondering if any of the tools that you shared with us today were helpful. Thank you. Thank you.

* * *

Original page: https://innovate-us.org/reading-out-loud-growing-strong-ai-tools-for-fluency-development · Licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org