/ecoz2-whale

ECOZ2 applied on whale song unit files

Primary LanguageShell

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 + fs / 1000, where the sampling frequency fs 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.

Some signal inspection

See signal-inspection.