Voice biometric is an easy and cost effective biometric technique which requires minimalistic hardware and software complexity. General voice biometric needs a voice phrase by user which is processed with Mel Filter and Vector Quantized features are extracted. Vector quantization reduces the codebook size but decreases the accuracy of recognition. Therefore we propose a voice biometric system where voice file's non quantized code books are matched with spoken phrase. In order to ensure security to such direct voice sample we embed the voice file in a randomly selected image using DWT technique. Imposters are exposed to only images and are unaware of the voice files. We show that the technique produces better efficiency in comparison to VQ based technique.