Real-time collaboration for Jupyter Notebooks, Linux Terminals, LaTeX, VS Code, R IDE, and more,
all in one place. Commercial Alternative to JupyterHub.
Real-time collaboration for Jupyter Notebooks, Linux Terminals, LaTeX, VS Code, R IDE, and more,
all in one place. Commercial Alternative to JupyterHub.
Path: blob/master/C5 - Sequence Models/Week 3/Trigger word detection/Trigger_word_detection_v2a.ipynb
Views: 4819
Trigger Word Detection
Welcome to the second and last programming assignment of Week 3!
In this week's videos, you learned about applying deep learning to speech recognition. In this assignment, you will construct a speech dataset and implement an algorithm for trigger word detection (sometimes also called keyword detection, or wake word detection).
Trigger word detection is the technology that allows devices like Amazon Alexa, Google Home, Apple Siri, and Baidu DuerOS to wake up upon hearing a certain word.
For this exercise, our trigger word will be "activate". Every time it hears you say "activate", it will make a "chiming" sound.
By the end of this assignment, you will be able to record a clip of yourself talking, and have the algorithm trigger a chime when it detects you saying "activate".
After completing this assignment, perhaps you can also extend it to run on your laptop so that every time you say "activate" it starts up your favorite app, or turns on a network connected lamp in your house, or triggers some other event?
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
Let's get started!
Table of Contents
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.
1.1 - 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.
Run the cells below to listen to some examples.
You will use these three types of recordings (positives/negatives/backgrounds) to create a labeled dataset.
1.2 - 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 (= ).
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.
Let's look at an example.
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 into the network, and so .
Now, you can define:
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.
You will use a Python module
pydubto synthesize audio, and it divides 10 seconds into 10,000 units.The output of our model will divide 10 seconds into 1,375 units.
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.
1.3 - 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 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.
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 represent whether or not someone has just finished saying "activate".
when that clip has finished saying "activate".
Given a background clip, we can initialize for all , since the clip doesn't contain any "activate".
When you insert or overlay an "activate" clip, you will also update labels for .
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".
Example
Suppose the synthesized "activate" clip ends at the 5 second mark in the 10 second audio - exactly halfway into the clip.
Recall that , so timestep
int(1375*0.5)corresponds to the moment 5 seconds into the audio clip.Set .
We will allow the GRU to detect "activate" anywhere within a short time-internal after this moment, so we actually set 50 consecutive values of the label to 1.
Specifically, we have .
Synthesized data is easier to label
This is another reason for synthesizing the training data: It's relatively straightforward to generate these labels as described above.
In contrast, if you have 10sec 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 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.
get_random_time_segment(segment_ms)Retrieves a random time segment from the background audio.
is_overlapping(segment_time, existing_segments)Checks if a time segment overlaps with existing segments
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_segmentandis_overlapping
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 durationsegment_ms.Please read through the code to make sure you understand what it is doing.
Check if audio clips are overlapping
Suppose you have inserted audio clips at segments (1000,1800) and (3400,4500).
The first segment starts at step 1000 and ends at step 1800.
The second segment starts at 3400 and ends at 4500.
If we are considering whether to insert a new audio clip at (3000,3600) does this overlap with one of the previously inserted segments?
In this case, (3000,3600) and (3400,4500) overlap, so we should decide against inserting a clip here.
For the purpose of this function, define (100,200) and (200,250) to be overlapping, since they overlap at timestep 200.
(100,199) and (200,250) are non-overlapping.
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.
You can use:
Hint: There is overlap if:
The new segment starts before the previous segment ends and
The new segment ends after the previous segment starts.
All tests passed!
Expected Output:
| **Overlap 1** | False |
| **Overlap 2** | True |
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.
(7286, 8201)
Timeouted
All tests passed!
Expected Output
| **Segment Time** | (2254, 3169) |
Insert ones for the labels of the positive target
Implement code to update the labels , assuming you just inserted an "activate" audio clip.
In the code below,
yis a(1,1375)dimensional vector, since .If the "activate" audio clip ends at time step , then set 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 arey[0][0]throughy[0][1374]because .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 (using a step discretization), we will use this formula:
All tests passed!
Expected Output
| **sanity checks**: | 0.0 1.0 0.0 |
Creating a training example
Finally, you can use insert_audio_clip and insert_ones to create a new training example.
Exercise 4 - create_training_example
Implement create_training_example(). You will need to carry out the following steps:
Initialize the label vector as a numpy array of zeros and shape .
Initialize the set of existing segments to an empty list.
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 .
Randomly select 0 to 2 negative audio clips, and insert them into the 10 second clip.
(2885, 3793)
(1726, 2450)
(6882, 7460)
All tests passed!
Expected Output 
Now you can listen to the training example you created and compare it to the spectrogram generated above.
Expected Output
Finally, you can plot the associated labels for the generated training example.
Expected Output 
You would like to save your dataset into a file that you can load later if you work in a more realistic environment. We let you the following code for reference. Don't try to run it into Coursera since the file system is read-only, and you cannot save files.
1.5 - 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.
2.1 - 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 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 .
Because 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 10sec 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.
Step 1: CONV layer. Use Conv1D() to implement this, with 196 filters, a filter size of 15 (kernel_size=15), and stride of 4. conv1d
Follow this with batch normalization. No parameters need to be set.
Follow this with a ReLu activation. Note that we can pass in the name of the desired activation as a string, all in lowercase letters.
Follow this with dropout, using a keep rate of 0.8
Step 2: First GRU layer. To generate the GRU layer, use 128 units.
Return sequences instead of just the last time step's prediction to ensure that all the GRU's hidden states are fed to the next layer.
Follow this with dropout, using a keep rate of 0.8.
Follow this with batch normalization. No parameters need to be set.
Step 3: Second GRU layer. This has the same specifications as the first GRU layer.
Follow this with a dropout, batch normalization, and then another dropout.
Step 4: Create a time-distributed dense layer as follows:
This creates a dense layer followed by a sigmoid, so that the parameters used for the dense layer are the same for every time step. Documentation:
To learn more, you can read this blog post How to Use the TimeDistributed Layer in Keras.
Exercise 5 - modelf
Implement modelf(), the architecture is presented in Figure 3.
All tests passed!
Let's print the model summary to keep track of the shapes.
Expected Output:
| **Total params** | 523,329 |
| **Trainable params** | 522,425 |
| **Non-trainable params** | 904 |
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.
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.
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.
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.
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.
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.
Let's explore how our model performs on two unseen audio clips from the development set. Lets first listen to the two dev set clips.
Now lets run the model on these audio clips and see if it adds a chime after "activate"!
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!
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.
Once you've uploaded your audio file to Coursera, put the path to your file in the variable below.
Finally, use the model to predict when you say activate in the 10 second audio clip, and trigger a chime. If beeps are not being added appropriately, try to adjust the chime_threshold.