Hi everyone, i want to ask does MIT app inventor integrate with a machine learning model for audio classification. ? for example i want to make ai models analyzes the student’s voice recoding and classifies the reading level..is there any easy way to make this ?
1 Like
This example
may help you with the classification.
The PAC (personal audio classifier extension) is downloaded here may also help
The ai part is up to you..
What have you tried @dd_dd
2 Likes
thanks for sharing.. for now, I’m still in the idea stage and haven’t started developing it yet. Sorry I want to ask again, how to implement the AI part for voice recording so that it can analyze students’ reading performance and identify their level of reading difficulties ?? I plan to build the app using MIT …. @SteveJG
Sí, es posible, pero MIT App Inventor no integra de forma nativa modelos avanzados de clasificación de audio. Una opción sencilla es entrenar un modelo con herramientas como Teachable Machine o TensorFlow y luego conectarlo a App Inventor mediante una API web. El estudiante graba su voz en la app, se envía el audio al modelo y este devuelve una clasificación (por ejemplo, nivel de lectura). Para prototipos simples funciona bien, aunque requiere algo de configuración externa.
i don’t understand. english please…
Use the small ‘globe’ to the left of the heart to implement translation to English
“Yes, it's possible, but MIT App Inventor doesn't natively integrate advanced audio classification models. A simple option is to train a model with tools such as Teachable Machine or TensorFlow and then connect it to App Inventor using a web API. The student records their voice in the app, the audio is sent to the model and the model returns a rating (for example, reading level). For simple prototypes it works well, although it requires some external configuration”
1 Like
Actually you seem to want to make an app to determine a person’s reading level (grade 3 English for example) 
My ai tells me (so don’t believe everything I am posting below) this:
AI Overview Using MIT App Inventor to build an AI app for reading assessment involves combining built-in audio components with text-matching blocks. The application records student audio, converts it into text, and compares it against a target passage to calculate reading accuracy, fluency, and identify errors.
Core Components Needed
To implement this application, you will need to drag the following components into your App Inventor project:
SpeechRecognizer: This non-visible component will take the live audio spoken by the student and convert it into text.
TextToSpeech: This component can read a target text out loud, or provide audible feedback to the student.
Clock (Optional): Used to time how long it takes a student to read the passage, giving you a WPM (words per minute) metric.
Label / TextBox: Displays the text and acts as an area to evaluate errors.
Implementation Logic and Blocks
Voice Recording & Transcription
The built-in SpeechRecognizer cannot directly process raw .mp3 or .wav files. Therefore, the student must read the passage into the microphone in real time.
Action: Use a "Record/Speak" button and call the SpeechRecognizer.GetText block.
Result: The AfterGettingText event will be triggered and output the student's spoken words as text.
Analyzing Reading Performance
Once you have the transcribed text, you can evaluate the student's performance by comparing it to the correct passage.
Accuracy: Use if/else logic to compare the student's recognized speech (the Result block) against a predefined list of words or the exact target passage. You can calculate the difference between the two strings to identify skipped words or mispronunciations
.Fluency: Use a Clock timer to measure the duration between when the student starts speaking and when they finish. You can calculate their reading pace by dividing the number of words by the time taken.
Identifying Reading Difficulties
You can code the app to identify specific reading struggles using algorithmic thresholds:Difficulty Level Adjustment: Present the student with progressively harder texts. If the app detects a word accuracy rate below a certain threshold (e.g., < 80%), the AI can assign an easier passage.
Hesitation & Pauses:
Use a Clock timer. If the time gap between individual words exceeds a set duration (e.g., 5 seconds), the app can classify this as a struggle or vocabulary difficulty.Recommended Audio Classifiers
If you want to move beyond basic speech-to-text and categorize specific types of reading struggles, you can utilize the Personal Audio Classifier (PAC) Extension available in App Inventor. This uses machine learning to classify unique 1-2 second audio bites. You can train an AI model to detect specific sounds (like a student stuttering, pausing, or struggling to sound out syllables) by training it via the in-browser PAC tool and exporting the model into your App Inventor workspace.
Will it work? I have no idea, you will have to try.
1 Like
aight.. noted the small globe
yes i wantt determine a person’s reading level…aight. thanks for suggestion and idea .. will try to do it 