This space captures VQ/HMM classification exercises on labelled whale song unit data.

Status

Complete exercises with some level of model tuning.

A general presentation about VQ/HMM and its application on the initial set of exercises described here can be found at https://github.com/ecoz2/ecoz2-doc/blob/master/exploration/vqhmm-whale.pdf.

Exercises

• exerc00: On 10 labelled files.

• exerc01: On a 4.5 hour recording.

• exerc02: As exerc01 but with some underlying adjustments in programs parameters and file organization.

• exerc03: related with the "windy" interval in the 4.5 recording:

  • exerc03a: excluding the windy interval
  • exerc03b: only considering the windy interval

• exerc04: As exerc01 but with number of refinements iterations set to 120

• exerc05: As exerc01 but with order of prediction P = 20 according to rule of thumb P = 4 + f_s / 1000, where the sampling frequency f_s is 16kHz in our case.

  • exerc05: Other parameters as in previous exercises
  • exerc05b: Only considering classes with at least 100 instances
  • exerc05c: Only considering classes with at least 200 instances

• exerc06: As exerc05c but merging classes "I" and "I2" into a "II" class.

  • exerc06

Some signal inspection

See signal-inspection.