All sound examples are monophonic and use a 22kHz sampling rate.
               Transforms use a 512-pt Hamming Window with 75% overlap.
            
                Example 1: Speech
             
                       Original

                    Reconstructions
                       Griffin & Lim,   4 iterations
                       Griffin & Lim,   100 iterations
                       RTISI,              4 iterations
                       RTISI_LA,       4 iterations

               Example 2: Popular Music

                       Original

                   Reconstructions
                       Griffin & Lim,   4 iterations
                       Griffin & Lim,   100 iterations
                       RTISI,              4 iterations
                       RTISI_LA,       4 iterations

     Time scale and pitch modifications using the RTSI_LA algorithm can be explored interactively.
     NOTE: You will need to be patient while the applet downloads - it includes speech and music sound files in wave format. This is also true the first time you use the slider to switch from the speech to the music example.
     Go to RTISI_LA interactive!




Examples of the Audio Test Segments used to derive averaged error measures in the paper:
Music -  chorus2, Music1, Music2, Music3, Music4, Music5
Speech - Female1, Male1