Speaker Recognition System

On-Line Speaker Recognition Module. By using this simple web form you can:

  • Add a speaker to database: just select the sound on your hard disk and specify ID (a non-negative integer number associated to speaker).
  • Recognize a speaker: just select the sound from your hard disk, code will return the ID of recognized speaker.
  • Remove database: all data will be deleted (inserted speaker and ID).
  • Code supports 8 and 16 bits WAV format. All sampling frequencies are supported.
Before starting to add new sounds to database it is recommended to delete any existing database. In this way you will be sure that database is initially empty: just click on "Remove Database" button.

We recommend to use for each speaker a sufficient number of training sounds, to improve recognition accuracy. Code has been tested with 239 speakers, 2419 sound files (an average number of 10 sound files for each speaker). The average duration of each sound file is 4.56 seconds, the sample rate is 48000 Hz and the number of bits per sample is 16. Using the Leave-One-Out method we have obtained an excellent recognition rate equal to 85.49% (Top-One recognition rate).

Web module to add a new speaker to database, an integer ID is required:

Sound to upload:

Insert speaker ID:

Web module to perform speaker recognition:

Sound to upload:

Web module to show database information:

Web module to remove any existing database:

In order to obtain the source code please email us luigi.rosa@tiscali.it.


October 30th, 2013


Text-independent speaker recognition based on Wavelet Transform. Click here to try this advanced script.


Face verification based on Advanced Correlation Filters. Click here to try this script developed entirely in PHP language.


PhiImage Image Processing Library. Click here for an on-line demo version.