This project demonstrates how I used AI to enhance student speech while preserving their own voice and ideas. I extracted a segment from a Zoom recording of an ESL lesson with a lower-intermediate student and created a “revised version” of our discussion using voice cloning. The goal was to let the student hear how they could express the same ideas with greater fluency and clearer pronunciation. Click here to listen to the Before Revision version and here to the After Revision version.
The process began with transcription and analysis. I recorded live one-on-one ESL sessions on Zoom and transcribed the audio using the Dictate feature in Microsoft Word. Once I had the written transcript, I imported it into NotebookLM, which identified specific areas where learners needed improvement in expressing their ideas. NotebookLM also highlighted the exact sections requiring attention by timestamp, allowing me to extract those parts and input them into ChatGPT.
When planning my lessons, I typically include a segment where students freely discuss a topic, such as a recent trip or their favorite activities. I extract this portion because it is where students most actively apply their learned vocabulary. I then use ChatGPT to rewrite the excerpt as if the learner were a native speaker. The point is to create a more natural and fluent version of the conversation. This provides a realistic model for the student to emulate.
To make the practice more engaging and authentic, I generated a realistic audio version of the revised conversation using Genny, an online AI voice generator. I chose Genny because it allows me to copy and paste the dialogue and automatically separates each line into individual exchanges, which saves significant editing time. Then, by cloning both my voice and the student’s voice, I was able to produce an audio dialogue that closely resembled what the student would sound like if speaking more fluently. This allowed learners to listen, imagine, and practice with an authentic-sounding model.
From the transcript, I also created a topic-related vocabulary table drawn directly from the lesson. Using NotebookLM, I scanned the transcript to identify each instance where a new word or phrase was introduced and then asked ChatGPT to transform these terms into a table with clear definitions and example sentences for the student to study. After each lesson, I would attach the AI-generated audio file, the revised transcript, and the vocabulary table into a handout file for the student to practice at home.
The project provided students with personalized, realistic language models they could listen to and practice. It also strengthened vocabulary retention through targeted exercises and increased learner confidence by allowing them to imagine how their speech could sound in authentic English contexts.
This project highlights my ability to conduct needs analysis using real learner data, apply AI tools to streamline content creation and personalization, and design microlearning workflows that reinforce practice and retention. It also demonstrates my skill in blending multimedia and instructional strategies to create engaging, learner-centered experiences.