Running CF-FM call segmentation

Here we will run the segment_and_measure function and store the results of how long each Cf/FM segment is.

Dataset creation

The synthetic dataset has already been created in a separate module. See ‘Generating the CF-FM synthetic calls’ in the main page.

It can take long

We’re running a few hundred synthetic audio clips with a few seconds (1-10s) needed per iteration. This could mean, it might take a while(5,10 or more minutes)!

import h5py
import itsfm
import pandas as pd
from tqdm import tqdm

Now, let’s load each synthetic call and proceed to save the results from the PWVD and peak-percentage based methods.

FM rate based segmentation

obtained = []

f = h5py.File('horseshoe_test.hdf5', 'r')
synthesised = pd.read_csv('horseshoe_test_parameters.csv')


fs = float(f['fs'][:])

parameters = {}
parameters['segment_method'] = 'pwvd'
parameters['window_size'] = int(fs*0.001)
parameters['fmrate_threshold'] = 2.0
parameters['max_acc'] = 10
parameters['extrap_window'] = 75*10**-6

for call_num in tqdm(range(synthesised.shape[0])):
    synthetic_call = f[str(call_num)][:]
    output = itsfm.segment_and_measure_call(synthetic_call, fs, **parameters)

    seg_output, call_parts, measurements= output
    # save the long format output into a wide format output to
    # allow comparison
    sub = measurements[['region_id', 'duration']]
    sub['call_number'] = call_num
    region_durations = sub.pivot(index='call_number',
                                 columns='region_id', values='duration')
    obtained.append(region_durations)

all_obtained = pd.concat(obtained)

all_obtained.to_csv('obtained_pwvd_horseshoe_sim.csv')

Peak-percentage based segmentation

pkpctg_parameters = {}
pkpctg_parameters['segment_method'] = 'peak_percentage'
pkpctg_parameters['peak_percentage'] = 0.99
pkpctg_parameters['window_size'] = 125
pkpctg_parameters['double_pass'] = True

pkpct_obtained = []

for call_num in tqdm(range(synthesised.shape[0])):
    synthetic_call = f[str(call_num)][:]
    output = itsfm.segment_and_measure_call(synthetic_call, fs, **pkpctg_parameters)

    seg_output, call_parts, measurements= output
    # save the long format output into a wide format output to
    # allow comparison
    sub = measurements[['region_id', 'duration']]
    sub['call_number'] = call_num
    region_durations = sub.pivot(index='call_number',
                                 columns='region_id', values='duration')
    pkpct_obtained.append(region_durations)


f.close()

pk_pctage = pd.concat(pkpct_obtained)

pk_pctage.to_csv('obtained_pkpct_horseshoe_sim.csv')