Application Pipeline#

This tutorial is available as an IPython notebook at malaya-speech/example/application-pipeline.

Use case#

  1. Read wav file.

  2. Apply noise reduction.

  3. Generate smaller frames for VAD. Read more about VAD at malaya-speech/example/vad.

  4. Detect VAD for each smaller frames.

  5. Visualize VAD.

  6. Group by VAD.

This is an application of malaya-speech Pipeline, read more about malaya-speech Pipeline at malaya-speech/example/pipeline.

[1]:

from malaya_speech import Pipeline
import malaya_speech
import numpy as np

[2]:

y, sr = malaya_speech.load('speech/podcast/example.wav')
len(y), sr

[2]:

(200160, 16000)

[3]:

vad = malaya_speech.vad.webrtc(sample_rate = sr, minimum_amplitude = int(np.quantile(np.abs(y), 0.2)))

Visualization pipeline#

[4]:

p = Pipeline()
frame = (
    p.map(malaya_speech.noise_reduction.reduce_noise_power)
    .map(malaya_speech.utils.astype.float_to_int)
    .map(malaya_speech.utils.generator.frames)
)
vad_map = frame.foreach_map(vad)
foreach = frame.foreach_zip(vad_map)
p.visualize()

[4]:
_images/load-application-pipeline_8_0.png 
[5]:

result = p(y)
result.keys()

[5]:

dict_keys(['reduce_noise_power', 'float_to_int', 'frames', 'vad', 'foreach_zip'])

[7]:

malaya_speech.extra.visualization.visualize_vad(y, result['foreach_zip'], sr)
_images/load-application-pipeline_10_0.png

Groupby pipeline#

[8]:

foreach.map(malaya_speech.group.group_frames)
p.visualize()

[8]:
_images/load-application-pipeline_12_0.png 
[9]:

result = p(y)

[10]:

result['group_frames']

[10]:

[(<malaya_speech.model.frame.Frame at 0x1470282d0>, True),
 (<malaya_speech.model.frame.Frame at 0x146fe4350>, False),
 (<malaya_speech.model.frame.Frame at 0x147028090>, True),
 (<malaya_speech.model.frame.Frame at 0x147028590>, False),
 (<malaya_speech.model.frame.Frame at 0x147028890>, True),
 (<malaya_speech.model.frame.Frame at 0x147028750>, False),
 (<malaya_speech.model.frame.Frame at 0x147026e10>, True),
 (<malaya_speech.model.frame.Frame at 0x147026d10>, False),
 (<malaya_speech.model.frame.Frame at 0x147026f10>, True),
 (<malaya_speech.model.frame.Frame at 0x147028c10>, False),
 (<malaya_speech.model.frame.Frame at 0x147026f50>, True),
 (<malaya_speech.model.frame.Frame at 0x147026d90>, False),
 (<malaya_speech.model.frame.Frame at 0x147026dd0>, True),
 (<malaya_speech.model.frame.Frame at 0x14702f310>, False),
 (<malaya_speech.model.frame.Frame at 0x14702f350>, True),
 (<malaya_speech.model.frame.Frame at 0x14702f3d0>, False),
 (<malaya_speech.model.frame.Frame at 0x14702f390>, True),
 (<malaya_speech.model.frame.Frame at 0x14702f410>, False),
 (<malaya_speech.model.frame.Frame at 0x14702f490>, True),
 (<malaya_speech.model.frame.Frame at 0x14696bc50>, False)]

[ ]: