Finding the acoustic features characterizing  basic speech sounds has been a debated mystery for many years. The  understanding of normal hearing listeners responses to nonsense  Consonant-Vowel (CV) sounds in noise is extremely limited.  Correlating confusion patterns and the acoustic cues available at a  given signal-to-noise ratio (SNR), using model neurograms, is the  key to this problem. We show that listeners use spectro-temporal  across-frequency timing cues to discriminate sounds within confusion  groups, e.g. that the recognition in noise of consonant /t/  critically depends on a ~20 ± 5 ms high frequency 3-8 kHz burst.  Adding masking noise can remove such features, leading to morphing,  turning one sound into another. Masking or truncating the burst  typically leads to the recognition of /p/ or /k/. Our analysis of  /ma/ and /na/ confusion patterns shows that an across-frequency  timing difference between high frequencies and mid-frequencies is  responsible for the discrimination between the two sounds. We will  play examples of plosives /p/, /t/, /k/ and  nasals /m/ and /n/, followed by several vowels, where the primary  consonant cue has been modified, leading to morphing.