Abstarct: Word boundary detection means the beginning and end of a word in a speech signal, which plays an important role for most applications such as speech recognition, text-to-speech system, and so on. In speech recognition system, most errors are due to incorrect recognition of word boundaries. Correct word boundaries reduces detection errors and improves systems performance.There are many algorithms used for detecting the boundary of words, some of which are not good enough to correct the word boundary due to the noise of the environment or a good pause between words. Which can be reduced by noise methods to be able to identify the borders well. While some methods are combined, they can better detection word boundaries. In this research, the aim is to detection word boundaries using time-frequency methods such as wavelet and Wigner-Ville. Time-frequency methods examine the signal in both time and frequency domains. Because the speech signal is an unstable signal and its characteristics change with time, the best way to analyze this signal is to extract features from those time-frequency methods. Different methods of extracting MFCC, PLP, LPC, wavelet analysis and Wigner-Ville method were examined to identify word boundaries. Two types of data were used to evaluate the results. In the TIMIT databaxse, time between words is very short and words are pronounced naturally and almost quickly, while in CADLab words are expressed numerically. The Wigner-Will method is able to identify word boundaries with 83.5% for the first data, and 68% for the second data is the best way to identify word boundaries. Entropy with 25% detection for the first data and ZRC with 30% detection for the second data has the lowest rate of correct detection of word boundaries among the methods performed.