AI-Powered Healthcare

AI-Powered Speech
Stuttering Detection

Upload your voice and receive intelligent speech fluency analysis using Machine Learning — powered by MFCC feature extraction and Random Forest classification.

8,000+
40
RF
High
Project Pipeline

How SpeakFlow AI Works

A complete ML pipeline from raw audio to intelligent prediction.

Audio Collection

8,000+ labeled audio samples of normal and stuttered speech collected from diverse speakers to build a robust training dataset.

Dataset

Audio Cleaning

Background noise removal, normalization, and silence trimming ensure clean signals before feature extraction begins.

Preprocessing

MFCC Extraction

Mel-Frequency Cepstral Coefficients (40 features) are extracted using Librosa — the gold standard for speech analysis.

Feature Engineering

Random Forest

An ensemble of decision trees classifies speech patterns, providing both prediction and probability estimates for each class.

ML Model

Prediction System

The final output delivers binary classification (Normal / Stutter) along with a confidence score and fluency rating.

Inference
Audio Analysis

Upload & Analyze Speech

Upload a WAV or MP3 file to run the stuttering detection model.

Drag & Drop Audio File

Supports WAV and MP3 formats

Analysis Details

  • MFCC features extracted (40 coefficients)
  • Random Forest ensemble inference
  • Confidence score provided
  • Speech fluency rating calculated
  • Result saved to history
  • PDF report available
WAV MP3
Live Recording

Record Your Voice

Use your microphone to record speech and analyze it directly.

00:00
AI Output

Results Dashboard

Your speech fluency analysis results appear here after processing.

No analysis run yet. Upload or record audio above to get results.

Detection History

Past Analyses

All previous analyses stored locally on your device.

# Date & Time File Name Result Confidence Action

No history yet. Run an analysis to see results here.

AI Pipeline

How the AI Works

A step-by-step walkthrough of the machine learning pipeline.

Step 1

Audio Upload

User uploads a WAV or MP3 audio file. The file is validated for format, size, and duration before being passed to the preprocessing pipeline.

Step 2

Audio Preprocessing

The audio signal is cleaned — background noise is reduced using spectral subtraction, amplitude is normalized, and silences are trimmed to isolate speech.

Step 3

MFCC Feature Extraction

Librosa extracts 40 Mel-Frequency Cepstral Coefficients from the processed signal. These features encode the tonal characteristics of speech, capturing rhythm disruptions indicative of stuttering.

Step 4

Random Forest Classification

The 40-feature vector is fed into a trained Random Forest ensemble. Each tree votes on the classification; the majority determines the predicted class and probability score.

Step 5

Prediction Result

The final prediction — Normal or Stuttering — is returned with a confidence score, fluency rating, and processing time. The result is stored in history and available as a PDF report.

By the Numbers

Project Statistics

0
Audio Samples
0
MFCC Features
0
Prediction Classes
0
Model Accuracy