def test_starting_info(self): self.assertIsNotNone( alysis.starting_info(x, alysis.pitch_detect(x, fs, chunk_size), fs, chunk_size))
from scipy.io import wavfile import pysptk import os from os.path import join import sys cwd = os.path.dirname(os.path.realpath(__file__)) sys.path.append(os.path.join(cwd, '..')) import ttsserver.espp.preprocess as prep import ttsserver.espp.analysis as alysis filename = os.path.join(cwd, "data/tmp.wav") fs, x = wavfile.read(filename) chunk_size = 1024 voice_samples = alysis.starting_info(x, alysis.pitch_detect(x, fs, chunk_size), fs, chunk_size)['VSamp'] class TestPreProcess(unittest.TestCase): def test_utterance_region_begin_samples(self): self.assertIsNotNone( prep.utterance_region_begin_samples(voice_samples)) def test_utterance_chunk(self): self.assertIsNotNone( prep.utterance_chunk( voice_samples, prep.utterance_region_begin_samples(voice_samples)[1])) def test_pre_process(self):
def test_pitch_detect(self): self.assertIsNotNone(alysis.pitch_detect(x, fs, chunk_size))