Context: The rapid advancement of deepfake technology has posed a significant challenge in the field of audio authentication, blurring the line between real and manipulated voices. Aims: This research aims to understand the intricate relationship between vocal feature parameters of authentic and deepfake voices and contribute to deepfake detection by analyzing vocal features and employing AI methodologies  Settings and Design: The study utilized the “DEEP-VOICE: DeepFake Voice Recognition” dataset and employed correlation analysis, descriptive statistics, visualization techniques and machine learning models. Methods and Material: Vocal features like duration, frequency, MFCCs and others were extracted. Correlation matrices, descriptive statistics and comparative visualizations were used to analyze differences between real and deepfake voices. Logistic Regression models were trained on the vocal features. Statistical analysis used: Correlation coefficients, means, standard deviations and machine learning model evaluation metrics were employed.  Results: Features like Duration, Number of Pulses and Voice Breaks exhibited negative correlations with deepfakes, while Mean Frequency and Median Frequency showed positive correlations with real voices. Descriptive statistics and visualizations highlighted differences in pitch and spectral characteristics. Logistic Regression models achieved promising performance, with pitch-related features being most effective.  Conclusions: The study identified key vocal feature parameters associated with real and deepfake voices, providing insights for developing robust voice authentication techniques and advancing deepfake detection.