본문 바로가기
COURSERA

week 3_Trigger Word Detection 실습 (Andrew Ng)

by HYUNHP 2022. 5. 17.
728x90
반응형

안녕하세요, HELLO

 

오늘은 DeepLearning.AI에서 진행하는 앤드류 응(Andrew Ng) 교수님의 딥러닝 전문화의 마지막이며, 다섯 번째 과정인 "Sequence Models"을 정리하려고 합니다.

 

"Sequence Models"의 강의를 통해 '시퀀스 모델과 음성 인식, 음악 합성, 챗봇, 기계 번역, 자연어 처리(NLP)' 등을 이해하고, 순환 신경망(RNN)과 GRU 및 LSTM, 트랜스포머 모델에 대해서 배우게 됩니다. 강의는 아래와 같이 구성되어 있습니다.

 

~ Recurrent Neural Networks

~ Natural Language Processing & Word Embeddings

~ Sequence Models & Attention Mechanism

~ Transformer Network

 

"Sequence Models" (Andrew Ng)의 3주차 "Trigger Word Detection"의 실습 내용입니다.

 

In this assignment you will learn to:

  • Structure a speech recognition project
  • Synthesize and process audio recordings to create train/dev datasets
  • Train a trigger word detection model and make predictions

CHAPTER 0. 'Packages'

 

CHAPTER 1. 'Data synthesis: Creating a Speech Dataset'

 

CHAPTER 2. 'The Model'

 

CHAPTER 3. 'Making Predictions'

 

CHAPTER 4. 'Try Your Own Example! (OPTIONAL/UNGRADED)'

CHAPTER 0. 'Packages'

 

import numpy as np
from pydub import AudioSegment
import random
import sys
import io
import os
import glob
import IPython
from td_utils import *
%matplotlib inline

CHAPTER 1. 'Data synthesis: Creating a Speech Dataset'

 

Let's start by building a dataset for your trigger word detection algorithm.

  • A speech dataset should ideally be as close as possible to the application you will want to run it on.
  • In this case, you'd like to detect the word "activate" in working environments (library, home, offices, open-spaces...).
  • Therefore, you need to create recordings with a mix of positive words ("activate") and negative words (random words other than activate) on different background sounds. Let's see how you can create such a dataset.

□ Listening to the Data

  • One of your friends is helping you out on this project, and they've gone to libraries, cafes, restaurants, homes and offices all around the region to record background noises, as well as snippets of audio of people saying positive/negative words. This dataset includes people speaking in a variety of accents.
  • In the raw_data directory, you can find a subset of the raw audio files of the positive words, negative words, and background noise. You will use these audio files to synthesize a dataset to train the model.
    • The "activate" directory contains positive examples of people saying the word "activate".
    • The "negatives" directory contains negative examples of people saying random words other than "activate".
    • There is one word per audio recording.
    • The "backgrounds" directory contains 10 second clips of background noise in different environments.

□ From Audio Recordings to Spectrograms

 

What really is an audio recording?

  • A microphone records little variations in air pressure over time, and it is these little variations in air pressure that your ear also perceives as sound.
  • You can think of an audio recording as a long list of numbers measuring the little air pressure changes detected by the microphone.
  • We will use audio sampled at 44100 Hz (or 44100 Hertz).
    • This means the microphone gives us 44,100 numbers per second.
    • Thus, a 10 second audio clip is represented by 441,000 numbers (= 10 ×44,10010 ×44,100).

Spectrogram

  • It is quite difficult to figure out from this "raw" representation of audio whether the word "activate" was said.
  • In order to help your sequence model more easily learn to detect trigger words, we will compute a spectrogram of the audio.
  • The spectrogram tells us how much different frequencies are present in an audio clip at any moment in time.
  • If you've ever taken an advanced class on signal processing or on Fourier transforms:
    • A spectrogram is computed by sliding a window over the raw audio signal, and calculating the most active frequencies in each window using a Fourier transform.
    • If you don't understand the previous sentence, don't worry about it.
x = graph_spectrogram("audio_examples/example_train.wav")

The graph above represents how active each frequency is (y axis) over a number of time-steps (x axis).

  • The color in the spectrogram shows the degree to which different frequencies are present (loud) in the audio at different points in time.
  • Green means a certain frequency is more active or more present in the audio clip (louder).
  • Blue squares denote less active frequencies.
  • The dimension of the output spectrogram depends upon the hyperparameters of the spectrogram software and the length of the input.
  • In this notebook, we will be working with 10 second audio clips as the "standard length" for our training examples.
    • The number of timesteps of the spectrogram will be 5511.
    • You'll see later that the spectrogram will be the input 𝑥x into the network, and so 𝑇𝑥=5511.

 

Tx = 5511 # The number of time steps input to the model from the spectrogram
n_freq = 101 # Number of frequencies input to the model at each time step of the spectrogram
Ty = 1375 # The number of time steps in the output of our model

 

Dividing into time-intervals

Note that we may divide a 10 second interval of time with different units (steps).

  • Raw audio divides 10 seconds into 441,000 units.
  • A spectrogram divides 10 seconds into 5,511 units.
    • 𝑇𝑥=5511T
  • You will use a Python module pydub to synthesize audio, and it divides 10 seconds into 10,000 units.
  • The output of our model will divide 10 seconds into 1,375 units.
    • 𝑇𝑦=1375
    • For each of the 1375 time steps, the model predicts whether someone recently finished saying the trigger word "activate".
  • All of these are hyperparameters and can be changed (except the 441000, which is a function of the microphone).
  • We have chosen values that are within the standard range used for speech systems.

□ Generating a Single Training Example

 

Benefits of synthesizing data

 

Because speech data is hard to acquire and label, you will synthesize your training data using the audio clips of activates, negatives, and backgrounds.

  • It is quite slow to record lots of 10 second audio clips with random "activates" in it.
  • Instead, it is easier to record lots of positives and negative words, and record background noise separately (or download background noise from free online sources).

Process for Synthesizing an audio clip

  • To synthesize a single training example, you will:
    • Pick a random 10 second background audio clip
    • Randomly insert 0-4 audio clips of "activate" into this 10 sec. clip
    • Randomly insert 0-2 audio clips of negative words into this 10 sec. clip
  • Because you had synthesized the word "activate" into the background clip, you know exactly when in the 10 second clip the "activate" makes its appearance.
    • You'll see later that this makes it easier to generate the labels 𝑦⟨𝑡⟩y⟨t⟩ as well.

Pydub

  • You will use the pydub package to manipulate audio.
  • Pydub converts raw audio files into lists of Pydub data structures.
    • Don't worry about the details of the data structures.
  • Pydub uses 1ms as the discretization interval (1 ms is 1 millisecond = 1/1000 seconds).
    • This is why a 10 second clip is always represented using 10,000 steps.
# Load audio segments using pydub 
activates, negatives, backgrounds = load_raw_audio('./raw_data/')

print("background len should be 10,000, since it is a 10 sec clip\n" + str(len(backgrounds[0])),"\n")
print("activate[0] len may be around 1000, since an `activate` audio clip is usually around 1 second (but varies a lot) \n" + str(len(activates[0])),"\n")
print("activate[1] len: different `activate` clips can have different lengths\n" + str(len(activates[1])),"\n")

 

Overlaying positive/negative 'word' audio clips on top of the background audio

  • Given a 10 second background clip and a short audio clip containing a positive or negative word, you need to be able to "add" the word audio clip on top of the background audio.
  • You will be inserting multiple clips of positive/negative words into the background, and you don't want to insert an "activate" or a random word somewhere that overlaps with another clip you had previously added.
    • To ensure that the 'word' audio segments do not overlap when inserted, you will keep track of the times of previously inserted audio clips.
  • To be clear, when you insert a 1 second "activate" onto a 10 second clip of cafe noise, you do not end up with an 11 sec clip.
    • The resulting audio clip is still 10 seconds long.
    • You'll see later how pydub allows you to do this.

 

Label the positive/negative words

  • Recall that the labels 𝑦⟨𝑡⟩ represent whether or not someone has just finished saying "activate".
    • 𝑦⟨𝑡⟩=1 when that clip has finished saying "activate".
    • Given a background clip, we can initialize 𝑦⟨𝑡⟩=0y⟨t⟩=0 for all 𝑡t, since the clip doesn't contain any "activate".
  • When you insert or overlay an "activate" clip, you will also update labels for 𝑦⟨𝑡⟩.
    • Rather than updating the label of a single time step, we will update 50 steps of the output to have target label 1.
    • Recall from the lecture on trigger word detection that updating several consecutive time steps can make the training data more balanced.
  • You will train a GRU (Gated Recurrent Unit) to detect when someone has finished saying "activate".

 

Synthesized data is easier to label

  • This is another reason for synthesizing the training data: It's relatively straightforward to generate these labels 𝑦⟨𝑡⟩ as described above.
  • In contrast, if you have 10 sec of audio recorded on a microphone, it's quite time consuming for a person to listen to it and mark manually exactly when "activate" finished.

Visualizing the labels

  • Here's a figure illustrating the labels 𝑦⟨𝑡⟩ in a clip.
    • We have inserted "activate", "innocent", "activate", "baby."
    • Note that the positive labels "1" are associated only with the positive words.

 

Helper functions

 

To implement the training set synthesis process, you will use the following helper functions.

  • All of these functions will use a 1ms discretization interval
  • The 10 seconds of audio is always discretized into 10,000 steps.
  1. get_random_time_segment(segment_ms)
    • Retrieves a random time segment from the background audio.
  2. is_overlapping(segment_time, existing_segments)
    • Checks if a time segment overlaps with existing segments
  3. insert_audio_clip(background, audio_clip, existing_times)
    • Inserts an audio segment at a random time in the background audio
    • Uses the functions get_random_time_segment and is_overlapping
  4. insert_ones(y, segment_end_ms)
    • Inserts additional 1's into the label vector y after the word "activate"

 

Get a random time segment

  • The function get_random_time_segment(segment_ms) returns a random time segment onto which we can insert an audio clip of duration segment_ms.
  • Please read through the code to make sure you understand what it is doing.
def get_random_time_segment(segment_ms):
    """
    Gets a random time segment of duration segment_ms in a 10,000 ms audio clip.
    
    Arguments:
    segment_ms -- the duration of the audio clip in ms ("ms" stands for "milliseconds")
    
    Returns:
    segment_time -- a tuple of (segment_start, segment_end) in ms
    """
    
    segment_start = np.random.randint(low=0, high=10000-segment_ms)   # Make sure segment doesn't run past the 10sec background 
    segment_end = segment_start + segment_ms - 1
    
    return (segment_start, segment_end)

□ Exercise 1 - is_overlapping

  • Implement is_overlapping(segment_time, existing_segments) to check if a new time segment overlaps with any of the previous segments.
  • You will need to carry out 2 steps:
  • Create a "False" flag, that you will later set to "True" if you find that there is an overlap.
  • Loop over the previous_segments' start and end times. Compare these times to the segment's start and end times. If there is an overlap, set the flag defined in (1) as True.

Hint: There is overlap if:

  • The new segment starts before the previous segment ends and
  • The new segment ends after the previous segment starts.
# UNQ_C1
# GRADED FUNCTION: is_overlapping

def is_overlapping(segment_time, previous_segments):
    """
    Checks if the time of a segment overlaps with the times of existing segments.
    
    Arguments:
    segment_time -- a tuple of (segment_start, segment_end) for the new segment
    previous_segments -- a list of tuples of (segment_start, segment_end) for the existing segments
    
    Returns:
    True if the time segment overlaps with any of the existing segments, False otherwise
    """
    
    segment_start, segment_end = segment_time
    
    ### START CODE HERE ### (≈ 4 lines)
    # Step 1: Initialize overlap as a "False" flag. (≈ 1 line)
    overlap = False
    
    # Step 2: loop over the previous_segments start and end times.
    # Compare start/end times and set the flag to True if there is an overlap (≈ 3 lines)
    for previous_start, previous_end in previous_segments: # @KEEP
        if segment_start <= previous_end and segment_end >= previous_start:
            overlap = True
            break
    ### END CODE HERE ###

    return overlap

 

# UNIT TEST
def is_overlapping_test(target):
    assert target((670, 1430), []) == False, "Overlap with an empty list must be False"
    assert target((500, 1000), [(100, 499), (1001, 1100)]) == False, "Almost overlap, but still False"
    assert target((750, 1250), [(100, 750), (1001, 1100)]) == True, "Must overlap with the end of first segment"
    assert target((750, 1250), [(300, 600), (1250, 1500)]) == True, "Must overlap with the begining of second segment"
    assert target((750, 1250), [(300, 600), (600, 1500), (1600, 1800)]) == True, "Is contained in second segment"
    assert target((800, 1100), [(300, 600), (900, 1000), (1600, 1800)]) == True, "New segment contains the second segment"

    print("\033[92m All tests passed!")
    
is_overlapping_test(is_overlapping)

 

overlap1 = is_overlapping((950, 1430), [(2000, 2550), (260, 949)])
overlap2 = is_overlapping((2305, 2950), [(824, 1532), (1900, 2305), (3424, 3656)])
print("Overlap 1 = ", overlap1)
print("Overlap 2 = ", overlap2)

 

Insert audio clip

  • Let's use the previous helper functions to insert a new audio clip onto the 10 second background at a random time.
  • We will ensure that any newly inserted segment doesn't overlap with previously inserted segments.

□ Exercise 2 - insert_audio_clip

  • Implement insert_audio_clip() to overlay an audio clip onto the background 10sec clip.
  • You implement 4 steps:
  • Get the length of the audio clip that is to be inserted.
    • Get a random time segment whose duration equals the duration of the audio clip that is to be inserted.
  • Make sure that the time segment does not overlap with any of the previous time segments.
    • If it is overlapping, then go back to step 1 and pick a new time segment.
  • Append the new time segment to the list of existing time segments
    • This keeps track of all the segments you've inserted.
  • Overlay the audio clip over the background using pydub. We have implemented this for you.

 

# UNQ_C2
# GRADED FUNCTION: insert_audio_clip

def insert_audio_clip(background, audio_clip, previous_segments):
    """
    Insert a new audio segment over the background noise at a random time step, ensuring that the 
    audio segment does not overlap with existing segments.
    
    Arguments:
    background -- a 10 second background audio recording.  
    audio_clip -- the audio clip to be inserted/overlaid. 
    previous_segments -- times where audio segments have already been placed
    
    Returns:
    new_background -- the updated background audio
    """
    
    # Get the duration of the audio clip in ms
    segment_ms = len(audio_clip)
    
    ### START CODE HERE ### 
    # Step 1: Use one of the helper functions to pick a random time segment onto which to insert 
    # the new audio clip. (≈ 1 line)
    segment_time = get_random_time_segment(segment_ms)
    
    # Step 2: Check if the new segment_time overlaps with one of the previous_segments. If so, keep 
    # picking new segment_time at random until it doesn't overlap. To avoid an endless loop
    # we retry 5 times(≈ 2 lines)
    retry = 5 # @KEEP 
    while is_overlapping(segment_time, previous_segments) is True and retry >= 0:
        segment_time = get_random_time_segment(segment_ms)
        retry = retry - 1
    ### END CODE HERE ###
        #print(segment_time)
    # if last try is not overlaping, insert it to the background
    if not is_overlapping(segment_time, previous_segments):
    ### START CODE HERE ### 
        # Step 3: Append the new segment_time to the list of previous_segments (≈ 1 line)
        previous_segments.append(segment_time)
    ### END CODE HERE ###
        # Step 4: Superpose audio segment and background
        new_background = background.overlay(audio_clip, position = segment_time[0])
    else:
        #print("Timeouted")
        new_background = background
        segment_time = (10000, 10000)
    
    return new_background, segment_time

 

# UNIT TEST
def insert_audio_clip_test(target):
    np.random.seed(5)
    audio_clip, segment_time = target(backgrounds[0], activates[0], [(0, 4400)])
    duration = segment_time[1] - segment_time[0]
    assert segment_time[0] > 4400, "Error: The audio clip is overlaping with the first segment"
    assert duration + 1 == len(activates[0]) , "The segment length must match the audio clip length"
    assert audio_clip != backgrounds[0] , "The audio clip must be different than the pure background"
    assert segment_time == (7286, 8201), f"Wrong segment. Expected: Expected: (7286, 8201) got:{segment_time}"

    # Not possible to insert clip into background
    audio_clip, segment_time = target(backgrounds[0], activates[0], [(0, 9999)])
    assert segment_time == (10000, 10000), "Segment must match the out by max-retry mark"
    assert audio_clip == backgrounds[0], "output audio clip must be exactly the same input background"

    print("\033[92m All tests passed!")

insert_audio_clip_test(insert_audio_clip)

 

np.random.seed(5)
audio_clip, segment_time = insert_audio_clip(backgrounds[0], activates[0], [(3790, 4400)])
audio_clip.export("insert_test.wav", format="wav")
print("Segment Time: ", segment_time)
IPython.display.Audio("insert_test.wav")


# Segment Time:  (2254, 3169)

 

Insert ones for the labels of the positive target

  • Implement code to update the labels 𝑦⟨𝑡⟩, assuming you just inserted an "activate" audio clip.
  • In the code below, y is a (1,1375) dimensional vector, since 𝑇𝑦=1375.
  • If the "activate" audio clip ends at time step 𝑡, then set 𝑦⟨𝑡+1⟩=1 and also set the next 49 additional consecutive values to 1.
    • Notice that if the target word appears near the end of the entire audio clip, there may not be 50 additional time steps to set to 1.
    • Make sure you don't run off the end of the array and try to update y[0][1375], since the valid indices are y[0][0] through y[0][1374] because 𝑇𝑦=1375Ty=1375.
    • So if "activate" ends at step 1370, you would get only set y[0][1371] = y[0][1372] = y[0][1373] = y[0][1374] = 1

□ Exercise 3 - insert_ones

 

Implement insert_ones().

  • You can use a for loop.
  • If you want to use Python's array slicing operations, you can do so as well.
  • If a segment ends at segment_end_ms (using a 10000 step discretization),
    • To convert it to the indexing for the outputs 𝑦y (using a 13751375 step discretization), we will use this formula:
    • segment_end_y = int(segment_end_ms * Ty / 10000.0)

 

# UNIT TEST
import random
def insert_ones_test(target):
    segment_end_y = random.randrange(0, Ty - 50) 
    segment_end_ms = int(segment_end_y * 10000.4) / Ty;    
    arr1 = target(np.zeros((1, Ty)), segment_end_ms)

    assert type(arr1) == np.ndarray, "Wrong type. Output must be a numpy array"
    assert arr1.shape == (1, Ty), "Wrong shape. It must match the input shape"
    assert np.sum(arr1) == 50, "It must insert exactly 50 ones"
    assert arr1[0][segment_end_y - 1] == 0, f"Array at {segment_end_y - 1} must be 0"
    assert arr1[0][segment_end_y] == 0, f"Array at {segment_end_y} must be 0"
    assert arr1[0][segment_end_y + 1] == 1, f"Array at {segment_end_y + 1} must be 1"
    assert arr1[0][segment_end_y + 50] == 1, f"Array at {segment_end_y + 50} must be 1"
    assert arr1[0][segment_end_y + 51] == 0, f"Array at {segment_end_y + 51} must be 0"

    print("\033[92m All tests passed!")
    
insert_ones_test(insert_ones)

 

arr1 = insert_ones(np.zeros((1, Ty)), 9700)
plt.plot(insert_ones(arr1, 4251)[0,:])
print("sanity checks:", arr1[0][1333], arr1[0][634], arr1[0][635])

□ Exercise 4 - create_training_example

 

Implement create_training_example(). You will need to carry out the following steps:

  1. Initialize the label vector 𝑦y as a numpy array of zeros and shape (1,𝑇𝑦).
  2. Initialize the set of existing segments to an empty list.
  3. Randomly select 0 to 4 "activate" audio clips, and insert them onto the 10 second clip. Also insert labels at the correct position in the label vector.
  4. Randomly select 0 to 2 negative audio clips, and insert them into the 10 second clip.

 

# UNQ_C4
# GRADED FUNCTION: create_training_example

def create_training_example(background, activates, negatives, Ty):
    """
    Creates a training example with a given background, activates, and negatives.
    
    Arguments:
    background -- a 10 second background audio recording
    activates -- a list of audio segments of the word "activate"
    negatives -- a list of audio segments of random words that are not "activate"
    Ty -- The number of time steps in the output

    Returns:
    x -- the spectrogram of the training example
    y -- the label at each time step of the spectrogram
    """
    
    # Make background quieter
    background = background - 20

    ### START CODE HERE ###
    # Step 1: Initialize y (label vector) of zeros (≈ 1 line)
    y = np.zeros((1, Ty))

    # Step 2: Initialize segment times as empty list (≈ 1 line)
    previous_segments = []
    ### END CODE HERE ###
    
    # Select 0-4 random "activate" audio clips from the entire list of "activates" recordings
    number_of_activates = np.random.randint(0, 5)
    random_indices = np.random.randint(len(activates), size=number_of_activates)
    random_activates = [activates[i] for i in random_indices]
    
    ### START CODE HERE ### (≈ 3 lines)
    # Step 3: Loop over randomly selected "activate" clips and insert in background
    for random_activate in random_activates: # @KEEP
        # Insert the audio clip on the background
        background, segment_time = insert_audio_clip(background, random_activate, previous_segments)
        # Retrieve segment_start and segment_end from segment_time
        segment_start, segment_end = segment_time
        # Insert labels in "y" at segment_end
        y = insert_ones(y, segment_end_ms = segment_end)
    ### END CODE HERE ###

    # Select 0-2 random negatives audio recordings from the entire list of "negatives" recordings
    number_of_negatives = np.random.randint(0, 3)
    random_indices = np.random.randint(len(negatives), size=number_of_negatives)
    random_negatives = [negatives[i] for i in random_indices]

    ### START CODE HERE ### (≈ 2 lines)
    # Step 4: Loop over randomly selected negative clips and insert in background
    for random_negative in random_negatives: # @KEEP
        # Insert the audio clip on the background 
        background, _ = insert_audio_clip(background, random_negative, previous_segments)
    ### END CODE HERE ###
    
    # Standardize the volume of the audio clip 
    background = match_target_amplitude(background, -20.0)

    # Export new training example 
    file_handle = background.export("train" + ".wav", format="wav")
    
    # Get and plot spectrogram of the new recording (background with superposition of positive and negatives)
    x = graph_spectrogram("train.wav")
    
    return x, y

 

# UNIT TEST
def create_training_example_test(target):
    np.random.seed(18)
    x, y = target(backgrounds[0], activates, negatives, 1375)
    
    assert type(x) == np.ndarray, "Wrong type for x"
    assert type(y) == np.ndarray, "Wrong type for y"
    assert tuple(x.shape) == (101, 5511), "Wrong shape for x"
    assert tuple(y.shape) == (1, 1375), "Wrong shape for y"
    assert np.all(x > 0), "All x values must be higher than 0"
    assert np.all(y >= 0), "All y values must be higher or equal than 0"
    assert np.all(y < 51), "All y values must be smaller than 51"
    assert np.sum(y) % 50 == 0, "Sum of activate marks must be a multiple of 50"
    assert np.isclose(np.linalg.norm(x), 39745552.52075), "Spectrogram is wrong. Check the parameters passed to the insert_audio_clip function"

    print("\033[92m All tests passed!")

create_training_example_test(create_training_example)

Finally, you can plot the associated labels for the generated training example.

□ Full Training Set

 

  • You've now implemented the code needed to generate a single training example.
  • We used this process to generate a large training set.
  • To save time, we generate a smaller training set of 32 examples.

 

# START SKIP FOR GRADING
np.random.seed(4543)
nsamples = 32
X = []
Y = []
for i in range(0, nsamples):
    if i%10 == 0:
        print(i)
    x, y = create_training_example(backgrounds[i % 2], activates, negatives, Ty)
    X.append(x.swapaxes(0,1))
    Y.append(y.swapaxes(0,1))
X = np.array(X)
Y = np.array(Y)
# END SKIP FOR GRADING

□ Development Set

 

  • To test our model, we recorded a development set of 25 examples.
  • While our training data is synthesized, we want to create a development set using the same distribution as the real inputs.
  • Thus, we recorded 25 10-second audio clips of people saying "activate" and other random words, and labeled them by hand.
  • This follows the principle described in Course 3 "Structuring Machine Learning Projects" that we should create the dev set to be as similar as possible to the test set distribution
    • This is why our dev set uses real audio rather than synthesized audio.

 

# Load preprocessed dev set examples
X_dev = np.load("./XY_dev/X_dev.npy")
Y_dev = np.load("./XY_dev/Y_dev.npy")

 

 

CHAPTER 2. 'The Model'

 

  • Now that you've built a dataset, let's write and train a trigger word detection model!
  • The model will use 1-D convolutional layers, GRU layers, and dense layers.
  • Let's load the packages that will allow you to use these layers in Keras. This might take a minute to load.

 

from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow.keras.models import Model, load_model, Sequential
from tensorflow.keras.layers import Dense, Activation, Dropout, Input, Masking, TimeDistributed, LSTM, Conv1D
from tensorflow.keras.layers import GRU, Bidirectional, BatchNormalization, Reshape
from tensorflow.keras.optimizers import Adam

□ Build the Model

 

Our goal is to build a network that will ingest a spectrogram and output a signal when it detects the trigger word. This network will use 4 layers:

* A convolutional layer
* Two GRU layers
* A dense layer.

Here is the architecture we will use.

 

1D convolutional layer

One key layer of this model is the 1D convolutional step (near the bottom of Figure 3).

  • It inputs the 5511 step spectrogram. Each step is a vector of 101 units.
  • It outputs a 1375 step output
  • This output is further processed by multiple layers to get the final 𝑇𝑦=1375Ty=1375 step output.
  • This 1D convolutional layer plays a role similar to the 2D convolutions you saw in Course 4, of extracting low-level features and then possibly generating an output of a smaller dimension.
  • Computationally, the 1-D conv layer also helps speed up the model because now the GRU can process only 1375 timesteps rather than 5511 timesteps.

 

GRU, dense and sigmoid

  • The two GRU layers read the sequence of inputs from left to right.
  • A dense plus sigmoid layer makes a prediction for 𝑦⟨𝑡⟩y⟨t⟩.
  • Because 𝑦y is a binary value (0 or 1), we use a sigmoid output at the last layer to estimate the chance of the output being 1, corresponding to the user having just said "activate".

 

Unidirectional RNN

  • Note that we use a unidirectional RNN rather than a bidirectional RNN.
  • This is really important for trigger word detection, since we want to be able to detect the trigger word almost immediately after it is said.
  • If we used a bidirectional RNN, we would have to wait for the whole 10 sec of audio to be recorded before we could tell if "activate" was said in the first second of the audio clip.

 

Implement the model

In the following model, the input of each layer is the output of the previous one. Implementing the model can be done in four steps.

 

□ modelf

 

Implement modelf(), the architecture is presented in the upper image.

 

# UNQ_C5
# GRADED FUNCTION: modelf

def modelf(input_shape):
    """
    Function creating the model's graph in Keras.
    
    Argument:
    input_shape -- shape of the model's input data (using Keras conventions)

    Returns:
    model -- Keras model instance
    """
    
    X_input = Input(shape = input_shape)
    
    ### START CODE HERE ###
    
    # Step 1: CONV layer (≈4 lines)
    # Add a Conv1D with 196 units, kernel size of 15 and stride of 4
    X = Conv1D(filters = 196, kernel_size = 15, strides = 4)(X_input)
    # Batch normalization
    X = BatchNormalization()(X)
    # ReLu activation
    X = Activation(activation = 'relu')(X)
    # dropout (use 0.8)
    X = Dropout(rate = 0.8)(X)                                  

    # Step 2: First GRU Layer (≈4 lines)
    # GRU (use 128 units and return the sequences)
    X = GRU(units = 128, return_sequences = True)(X)
    # dropout (use 0.8)
    X = Dropout(rate = 0.8)(X)
    # Batch normalization.
    X = BatchNormalization()(X)                           
    
    # Step 3: Second GRU Layer (≈4 lines)
    # GRU (use 128 units and return the sequences)
    X = GRU(units = 128, return_sequences = True)(X)
    # dropout (use 0.8)
    X = Dropout(rate = 0.8)(X)       
    # Batch normalization
    X = BatchNormalization()(X)   
    # dropout (use 0.8)
    X = Dropout(rate = 0.8)(X)                                  
    
    # Step 4: Time-distributed dense layer (≈1 line)
    # TimeDistributed  with sigmoid activation 
    X = TimeDistributed(Dense(1, activation = 'sigmoid'))(X) 

    ### END CODE HERE ###

    model = Model(inputs = X_input, outputs = X)
    
    return model

 

# UNIT TEST
from test_utils import *

def modelf_test(target):
    Tx = 5511
    n_freq = 101
    model = target(input_shape = (Tx, n_freq))
    expected_model = [['InputLayer', [(None, 5511, 101)], 0],
                     ['Conv1D', (None, 1375, 196), 297136, 'valid', 'linear', (4,), (15,), 'GlorotUniform'],
                     ['BatchNormalization', (None, 1375, 196), 784],
                     ['Activation', (None, 1375, 196), 0],
                     ['Dropout', (None, 1375, 196), 0, 0.8],
                     ['GRU', (None, 1375, 128), 125184, True],
                     ['Dropout', (None, 1375, 128), 0, 0.8],
                     ['BatchNormalization', (None, 1375, 128), 512],
                     ['GRU', (None, 1375, 128), 99072, True],
                     ['Dropout', (None, 1375, 128), 0, 0.8],
                     ['BatchNormalization', (None, 1375, 128), 512],
                     ['Dropout', (None, 1375, 128), 0, 0.8],
                     ['TimeDistributed', (None, 1375, 1), 129, 'sigmoid']]
    comparator(summary(model), expected_model)
    
    
modelf_test(modelf)

 

model = modelf(input_shape = (Tx, n_freq))
model.summary()

The output of the network is of shape (None, 1375, 1) while the input is (None, 5511, 101). The Conv1D has reduced the number of steps from 5511 to 1375. 

□ Fit the Model

 

  • Trigger word detection takes a long time to train.
  • To save time, we've already trained a model for about 3 hours on a GPU using the architecture you built above, and a large training set of about 4000 examples.
  • Let's load the model.

 

from tensorflow.keras.models import model_from_json

json_file = open('./models/model.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
model = model_from_json(loaded_model_json)
model.load_weights('./models/model.h5')

■ Block Training for BatchNormalization Layers

 

If you are going to fine-tune a pretrained model, it is important that you block the weights of all your batchnormalization layers. If you are going to train a new model from scratch, skip the next cell.

 

model.layers[2].trainable = False
model.layers[7].trainable = False
model.layers[10].trainable = False

 

You can train the model further, using the Adam optimizer and binary cross entropy loss, as follows. This will run quickly because we are training just for two epochs and with a small training set of 32 examples. 

 

opt = Adam(lr=1e-6, beta_1=0.9, beta_2=0.999)
model.compile(loss='binary_crossentropy', optimizer=opt, metrics=["accuracy"])
model.fit(X, Y, batch_size = 16, epochs=1)
2/2 [==============================] - 4s 2s/step - loss: 0.1731 - accuracy: 0.9455

□ Test the Model

 

Finally, let's see how your model performs on the dev set.

 

loss, acc, = model.evaluate(X_dev, Y_dev)
print("Dev set accuracy = ", acc)

 

1/1 [==============================] - 0s 1ms/step - loss: 0.1887 - accuracy: 0.9239
Dev set accuracy =  0.9238690733909607
 

This looks pretty good!

  • However, accuracy isn't a great metric for this task
    • Since the labels are heavily skewed to 0's, a neural network that just outputs 0's would get slightly over 90% accuracy.
  • We could define more useful metrics such as F1 score or Precision/Recall.
    • Let's not bother with that here, and instead just empirically see how the model does with some predictions.

 

반응형

 

CHAPTER 3. 'Making Predictions'

 

Now that you have built a working model for trigger word detection, let's use it to make predictions. This code snippet runs audio (saved in a wav file) through the network.

 

def detect_triggerword(filename):
    plt.subplot(2, 1, 1)
    
    # Correct the amplitude of the input file before prediction 
    audio_clip = AudioSegment.from_wav(filename)
    audio_clip = match_target_amplitude(audio_clip, -20.0)
    file_handle = audio_clip.export("tmp.wav", format="wav")
    filename = "tmp.wav"

    x = graph_spectrogram(filename)
    # the spectrogram outputs (freqs, Tx) and we want (Tx, freqs) to input into the model
    x  = x.swapaxes(0,1)
    x = np.expand_dims(x, axis=0)
    predictions = model.predict(x)
    
    plt.subplot(2, 1, 2)
    plt.plot(predictions[0,:,0])
    plt.ylabel('probability')
    plt.show()
    return predictions

 

Insert a chime to acknowledge the "activate" trigger

  • Once you've estimated the probability of having detected the word "activate" at each output step, you can trigger a "chiming" sound to play when the probability is above a certain threshold.
  • 𝑦⟨𝑡⟩ might be near 1 for many values in a row after "activate" is said, yet we want to chime only once.
    • So we will insert a chime sound at most once every 75 output steps.
    • This will help prevent us from inserting two chimes for a single instance of "activate".
    • This plays a role similar to non-max suppression from computer vision.

 

chime_file = "audio_examples/chime.wav"
def chime_on_activate(filename, predictions, threshold):
    audio_clip = AudioSegment.from_wav(filename)
    chime = AudioSegment.from_wav(chime_file)
    Ty = predictions.shape[1]
    # Step 1: Initialize the number of consecutive output steps to 0
    consecutive_timesteps = 0
    # Step 2: Loop over the output steps in the y
    for i in range(Ty):
        # Step 3: Increment consecutive output steps
        consecutive_timesteps += 1
        # Step 4: If prediction is higher than the threshold and more than 20 consecutive output steps have passed
        if consecutive_timesteps > 20:
            # Step 5: Superpose audio and background using pydub
            audio_clip = audio_clip.overlay(chime, position = ((i / Ty) * audio_clip.duration_seconds) * 1000)
            # Step 6: Reset consecutive output steps to 0
            consecutive_timesteps = 0
        # if amplitude is smaller than the threshold reset the consecutive_timesteps counter
        if predictions[0, i, 0] < threshold:
            consecutive_timesteps = 0
        
    audio_clip.export("chime_output.wav", format='wav')

□ Test on Dev Examples

 

Now lets run the model on these audio clips and see if it adds a chime after "activate"!

 

filename = "./raw_data/dev/1.wav"
prediction = detect_triggerword(filename)
chime_on_activate(filename, prediction, 0.5)
IPython.display.Audio("./chime_output.wav")

Congratulations

You've come to the end of this assignment!

Here's what you should remember:

  • Data synthesis is an effective way to create a large training set for speech problems, specifically trigger word detection.
  • Using a spectrogram and optionally a 1D conv layer is a common pre-processing step prior to passing audio data to an RNN, GRU or LSTM.
  • An end-to-end deep learning approach can be used to build a very effective trigger word detection system.

Congratulations on finishing this assignment!

 

 

CHAPTER 4. 'Try Your Own Example! (OPTIONAL/UNGRADED)'

 

In this optional and ungraded portion of this notebook, you can try your model on your own audio clips!

  • Record a 10 second audio clip of you saying the word "activate" and other random words, and upload it to the Coursera hub as myaudio.wav.
  • Be sure to upload the audio as a wav file.
  • If your audio is recorded in a different format (such as mp3) there is free software that you can find online for converting it to wav.
  • If your audio recording is not 10 seconds, the code below will either trim or pad it as needed to make it 10 seconds.

 

# Preprocess the audio to the correct format
def preprocess_audio(filename):
    # Trim or pad audio segment to 10000ms
    padding = AudioSegment.silent(duration=10000)
    segment = AudioSegment.from_wav(filename)[:10000]
    segment = padding.overlay(segment)
    # Set frame rate to 44100
    segment = segment.set_frame_rate(44100)
    # Export as wav
    segment.export(filename, format='wav')

 

your_filename = "audio_examples/my_audio.wav"
chime_threshold = 0.5
prediction = detect_triggerword(your_filename)
chime_on_activate(your_filename, prediction, chime_threshold)
IPython.display.Audio("./chime_output.wav")

■ 마무리

 

"Sequence Models" (Andrew Ng)의 3주차 "Trigger Word Detection"의 실습에 대해서 정리해봤습니다.

 

그럼 오늘 하루도 즐거운 나날 되길 기도하겠습니다

좋아요와 댓글 부탁드립니다 :)

 

감사합니다.

 

반응형
저작자표시 비영리 동일조건

'COURSERA' 카테고리의 다른 글

week 4_Transformer Network 연습 문제 (Andrew Ng)  (0) 2022.05.17
week 4_Transformer Network (Andrew Ng)  (0) 2022.05.17
week 3_Neural Machine Translation 실습 (Andrew Ng)  (0) 2022.05.17
week 3_Sequence Models & Attention Mechanism 연습 문제 (Andrew Ng)  (0) 2022.05.17
week 3_Sequence Models & Attention Mechanism (Andrew Ng)  (1) 2022.05.10

관련글

댓글

티스토리툴바