Preprocessing and core selection

Basic info

• Float reference:
  • Encoder (float): 6432cc34
• Fixed point:
  • Encoder (fixed): 24cb7eef

Description

We have run the encoder on different database and compared the floating point to fixed point core selection for different configurations. This ticket is just to keep track of the results obtained.

In general, the first stage speech/music selection between float and fix seems to work well and have consistent number when compared to the EVS float/fix performances. On the other hand, the 2nd and 3rd level (sp_aud_decision1 and sp_aud_decision2) show a more deviation which lead to a greater deviation of the core selection as well.

Vad_flag and localVAD_HE_SAD seems pretty accurate so far.

@malenovsky, let me know you impression here.