Profanities are censored due to their possible negative effects on social media. However, these are purposefully obfuscated in texts and remain undetected by models. In addition, obfuscation by phonetic similarity remains a concern in various existing works. Therefore, this study aimed to address this issue by utilizing text-to-speech (TTS) to classify the obfuscated word based on its phonetic information. In line with this, a dataset with 8,190 obfuscated profanities was converted to audio using TTS and underwent Short-Time Fourier Transform to amplify vowels, thereby enhancing speech intelligibility. Mel-frequency Cepstral Coefficients are extracted in the signals. For classification, a Convolutional Neural Network (CNN) was used. The results revealed that the CNN model recorded an accuracy score of 97.74%. On the other hand, an average Euclidean distance of 0.08 and a Cosine distance of 0.0003 indicated that vowel amplification provided little significance in the model’s performance due to the slight difference observed on the preprocessed signal. As an outcome of this experiment, text-to-speech is suggested to be utilized for other natural language processing tasks, such as the detection of profanities in text. However, further studies on other methods of enhancing vowel sounds are to be considered to improve speech intelligibility and facilitate better correction of obfuscated profanities.