A first grade student is trying to read a passage on her iPad. A digital avatar Amira, clad in olive green, is listening. Her face isn’t particularly demonstrative, but she’s trying her best with emphatic pats-on-the-back when the student gets something right. When the reader skips a word, or mispronounces it, Amira displays the kind of dispassionate instruction that only artificially created avatars can.
“Keep going,” Amira says, softly.
Amira is the invention of Amira Learning, a six-year-old edtech company that fuses voice-based artificial intelligence into reading activities, guided by an eponymous AI bot. Amira Learning is only one of a slew of edtech companies that have leveraged the advancements in voice-based AI to help improve foundational reading skills for learners from kindergarten right up to fourth grade.
These systems act as guides for students, and as they read a text, analyze their speech to identify the proficiency level of the reader. They try to replicate the experience of a teacher listening carefully and identifying potential problem areas in comprehension, pronunciation and letter recognition. Voice technology — especially the use of an AI bot that talks back to the learner — has injected reading practice with the kind of feedback that was only possible with one-on-one tutoring before.
School district leaders have taken note, developing multi-year adoption plans for their schools. At Savannah-Chatham County Public Schools (SCCPSS) in Georgia, a three-school pilot in January 2020 grew into a district-wide online and offline reading program for students across its 34 elementary schools, according to Andrea Burkiett, director of curriculum and instruction for the district.
“Between May and June of 2020, our students had collectively read 77,000 minutes on the Amira platform. During COVID, we weren’t sure what other literacy instruction they were able to get. So we were pretty happy with Amira,” Burkiett says.
When schools reopened in 2021, Burkiett and her team decided to roll out the program across the district for students in kindergarten to third grade.
“The recording feature was very advantageous because it allowed teachers to listen to students reading, even if they weren’t physically present in schools,” Burkiett says.
There’s a growing market for the convenience, accurate feedback and interaction that voice-based AI reading tools provide, due to the unprecedented fall in NAEP reading scores across the country coupled with mounting concerns about modern strategies for teaching literacy. There’s also a need for tools that cater to varying reading levels within the same classroom and give teachers quick feedback on which students need most attention, and in which area.
Diagnosing dyslexia early, for instance.
“What voice tech allows you to do, at scale, is to make an earlier diagnosis. The challenge with screening for dyslexia is that if you’re only dependent on a human, you’re limited in terms of the resource,” says Martyn Farrows, CEO of SoapBox Labs, a voice tech company.
SoapBox Labs has created its own patented voice recognition tool that’s become an off-the-shelf API product for literacy apps like EarlyBird and Amplify. Other companies with products in the market are Bamboo Learning, Ello and Imagine Learning.
The appeal of voice technology as a screening tool, a method for playful reading practice and a strategy for offering feedback is clear. That’s why Scholastic, a leader in education products across the country, has inked what the company calls a “long-term” deal with SoapBox to launch a new multimedia phonics curriculum called Ready4Reading for students in kindergarten through third grade. With potential access to Scholastic’s 115,000-school network, voice tech tools like SoapBox can quickly become mainstream for reading literacy programs.
The Power of Voice
From an instructional point of view, voice tech seems to cover two important bases — it’s interactive, and it’s able to act as a “wingman” to teachers in a classroom.
For example, if a student reads a sentence like “I met the president,” the AI will pay attention to various parameters of each word in the sentence at the same time. The voice bot jumps in with immediate feedback at the phonemic level if the student isn’t quite linking the right sounds with the right letters. At the backend, the teacher sees a visual explanation of each word read right and each word skipped or mispronounced.
“Lexile levels, words read per minute, fluency, don’t mean anything to students,” Burkiett says. What does? “It’s the immediate engagement of saying, OK, you read that wrong, let’s try it again, let’s practice.”
Instant feedback is also a result of how well these voice recognition tools are able to pick up differing accents, ambient sounds and speaking styles. It’s what Megan Van Fossan calls the “lack of bias.”
Van Fossan is the superintendent at the Sto-Rox public school district in Allegheny County, Pennsylvania. Sto-Rox schools use an early literacy screener called EarlyBird, whose underlying voice tech is powered by SoapBox’s API, for kindergarten and first grade students.
“I was looking for something that was backed by great research, had good tech support and would have professional development,” Van Fossan says.
The lack of bias that appeals to Van Fossan is a deliberate strategy by SoapBox to build a more inclusive and representative databank of voices.
“With AI and machine learning technologies, often what will happen is that they will take a pre-existing dataset, and then try to build the product on top of that. The challenge with that approach is that often that pre-existing dataset can have already inherent biases built into it, or it may not be representative. So we made sure that our dataset is representative of children from all backgrounds, accents, dialects.”
Kindergarten teachers at Sto-Rox transitioned from traditional assessments like DIBELS to EarlyBird at the start of the 2022-2023 school year. Teachers in the district, like Sherine Raymond, believe that EarlyBird offers a more detailed picture of what level students are reading at. DIBELS is a basic scorecard, says Raymond, while EarlyBird goes into granular detailed feedback about phonemic-level awareness.
“I can see where the kids are substituting words, what their deficiencies are,” Raymond sys.
An EarlyBird assessment takes about 45 minutes per child to administer, and Van Fossan says it’s done at the beginning, middle and end of every school year. The assessments, in addition to giving feedback about every student, have also led to some structural changes in how the Sto-Rox district’s kindergarten sections are organized.
“When we got the data back [from the assessments], we were alarmed to see that students were not reading at the level they should be. They didn’t recognize letters they should be,” Raymond says.
The data instigated a conversation about how kindergarten was compartmentalized: Raymond would teach reading, while another teacher would do math. What this led to was insufficient time to practice things like specific reading techniques — on top of trying to teach children how to engage in basic social behaviors.
“During COVID, kindergarten students never gained pre-literacy skills like standing in line, or sitting in a chair. And then we were asking them to switch classes,” Raymond says.
The compartments have dissolved now, across the district, putting teachers like Raymond into self-contained classrooms.
What Difference Does Voice Tech Make?
For young readers in Georgia, voice technology is already prompting improvements, according to SCCPSS leaders. Burkiett explains Amira’s impact by pointing to progress for children in the lowest-scoring percentile when it comes to reading fluency.
“We reduced the number of students who were at or below the 25th percentile by 7 percent,” Burkiett says.
The one-on-one reading lessons came to SCCPSS schools as the district was adopting a one-to-one device program. Burkiett says the cost to the district, over the last three years of using Amira, amounts to a ballpark figure of $176 per child annually. The district is going in for a two-year renewal this year, she adds.
Meanwhile, SoapBox’s website claims high efficacy for its patented speech recognition technology. For instance, in partnership with Amplify, SoapBox ran a beta test that the company says found a 96 percent correlation between its own automated assessments and how a human would score in its place.
SoapBox launched a multi-year partnership with the Florida Center for Reading Research at Florida State University in 2019, to develop literacy assessments, according to the center’s website. The partnership also included “pilot studies” of SoapBox’s speech recognition technology by assessing students in kindergarten to second grade from several states.
In 2022, Bamboo Learning commissioned a third-party assessment of its learning platform with 82 first grade students. The study found that, on average, students started out reading books intended for kindergarten and first grade. After six weeks of using the platform, they were reading books designed for first and second grade. Bamboo Learning’s study also shows an improvement in attitudes toward reading.
In addition to aiding students, voice tech has also eased the time burden for educators, according to the teachers that EdSurge spoke with. Reading activities that ordinarily would take over an hour can now generate real-time feedback within minutes. Previously, teachers had to first spend time listening to every student and design interventions according to their reading levels. Now, as Burkiett puts it, the feedback and the intervention are available at the touch of a button. A teacher can also spend more time with slower readers, and let the more advanced groups practice on their own.
There’s something to be said about such automation. Adopting voice tech, layered with the easy-to-read dashboards and in-built practice, can help teachers get to the point of innovation, instead of doing each step manually. But both Van Fossan and Burkiett agree that the granular feedback could feel overwhelming for teachers.
“Teachers can spend all their time going over each recording, so we’re training them to do spot checks,” says Burkiett.
The SCCPSS district has implemented a monthly meeting for teachers to figure out which piece of data they want to work with. The recordings indicate which words were read correctly (highlighted green), which were wrong (highlighted red), and which weren’t recorded properly (in yellow). Teachers can go back to the recording to listen for the “yellow” words, and figure out if the student read it correctly or not. Teachers can’t avoid looking at the data, so they have to be strategic about it, Burkiett says.
At Sto-Rox, reading teachers go through 45 minutes of professional development every day, which includes time to review student data and complete training offered by the AIM Institute of Learning and Research.
“If you look at schools in Norway, Singapore or Sweden, they give an amazing amount of time for educators to collaborate and learn,” Van Fossan says. “Teachers need time to operationalize their learning.”
Beyond Sto-Rox and SCCPSS, the New York and Chicago public school systems have also shown interest in using voice recognition tech to improve their literacy levels. Irina Fine, CEO of Bamboo Learning, has also struck partnerships with nonprofit organizations that are willing to underwrite the use of their products by underserved school communities.
But just as much as these tools might alter reading interventions in schools, their spread will also change the DNA of these tools.
Farrows, of SoapBox Labs, says that more teachers have begun focusing on phonemic-level awareness. They also want to know how students are intonating, which indicates whether students understand what they’re reading. Teachers using Bamboo Learning want to focus on both open-ended responses (learners saying entire sentences) and on true/false type of statements, according to Fine.
Teachers have a finite amount of time in a day to spend on reading instruction. So it’s helpful if each voice-based edtech company offers every tool these educators might need. This also means that with a couple of years of feedback, all tools could mirror each other in terms of usability.
How will school districts choose among products, then? Fine says that Bamboo Learning has built out an assessment platform too, in addition to its reading products, so that schools trying to decide how to spend their money and time on reading tools can assess impact as they roll out the program in their district. “We’re even willing to offer the program at half the cost, if the school lets us do an assessment,” Fine says. Right now, the program costs $36 per child. Bamboo’s willing to offer it at $18.
Beyond offering discounts, edtech founders like Fine are displaying their keenness to districts in other ways, too.
“I think only 15 percent of the federal funding from COVID allocated has been spent. So there’s money there,” Fine says. “When we go into pilot, we’re going to do everything, set up the teacher accounts, roster, etc., so they don’t have to spend any additional time with the technology.”