Posit AI Weblog: Audio classification with torch

Variations on a theme

Easy audio classification with Keras, Audio classification with Keras: Wanting nearer on the non-deep studying components, Easy audio classification with torch: No, this isn’t the primary put up on this weblog that introduces speech classification utilizing deep studying. With two of these posts (the “utilized” ones) it shares the final setup, the kind of deep-learning structure employed, and the dataset used. With the third, it has in frequent the curiosity within the concepts and ideas concerned. Every of those posts has a unique focus – must you learn this one?

Properly, in fact I can’t say “no” – all of the extra so as a result of, right here, you’ve an abbreviated and condensed model of the chapter on this matter within the forthcoming e book from CRC Press, Deep Studying and Scientific Computing with R torch. By the use of comparability with the earlier put up that used torch, written by the creator and maintainer of torchaudio, Athos Damiani, important developments have taken place within the torch ecosystem, the tip end result being that the code received loads simpler (particularly within the mannequin coaching half). That stated, let’s finish the preamble already, and plunge into the subject!

Inspecting the information

We use the speech instructions dataset (Warden (2018)) that comes with torchaudio. The dataset holds recordings of thirty completely different one- or two-syllable phrases, uttered by completely different audio system. There are about 65,000 audio information total. Our process will likely be to foretell, from the audio solely, which of thirty attainable phrases was pronounced.

library(torch)
library(torchaudio)
library(luz)

ds <- speechcommand_dataset(
  root = "~/.torch-datasets", 
  url = "speech_commands_v0.01",
  obtain = TRUE
)

We begin by inspecting the information.

[1]  "mattress"    "chicken"   "cat"    "canine"    "down"   "eight"
[7]  "5"   "4"   "go"     "completely happy"  "home"  "left"
[32] " marvin" "9"   "no"     "off"    "on"     "one"
[19] "proper"  "seven" "sheila" "six"    "cease"   "three"
[25]  "tree"   "two"    "up"     "wow"    "sure"    "zero" 

Choosing a pattern at random, we see that the data we’ll want is contained in 4 properties: waveform, sample_rate, label_index, and label.

The primary, waveform, will likely be our predictor.

pattern <- ds[2000]
dim(pattern$waveform)

[1]     1 16000

Particular person tensor values are centered at zero, and vary between -1 and 1. There are 16,000 of them, reflecting the truth that the recording lasted for one second, and was registered at (or has been transformed to, by the dataset creators) a charge of 16,000 samples per second. The latter info is saved in pattern$sample_rate:

[1] 16000

All recordings have been sampled on the similar charge. Their size nearly at all times equals one second; the – very – few sounds which can be minimally longer we will safely truncate.

Lastly, the goal is saved, in integer type, in pattern$label_index, the corresponding phrase being obtainable from pattern$label:

pattern$label
pattern$label_index

[1] "chicken"
torch_tensor
2
[ CPULongType{} ]

How does this audio sign “look?”

library(ggplot2)

df <- information.body(
  x = 1:size(pattern$waveform[1]),
  y = as.numeric(pattern$waveform[1])
  )

ggplot(df, aes(x = x, y = y)) +
  geom_line(measurement = 0.3) +
  ggtitle(
    paste0(
      "The spoken phrase "", pattern$label, "": Sound wave"
    )
  ) +
  xlab("time") +
  ylab("amplitude") +
  theme_minimal()

What we see is a sequence of amplitudes, reflecting the sound wave produced by somebody saying “chicken.” Put in a different way, we now have right here a time collection of “loudness values.” Even for consultants, guessing which phrase resulted in these amplitudes is an unimaginable process. That is the place area information is available in. The skilled could not have the ability to make a lot of the sign on this illustration; however they could know a method to extra meaningfully symbolize it.

Two equal representations

Think about that as an alternative of as a sequence of amplitudes over time, the above wave had been represented in a method that had no details about time in any respect. Subsequent, think about we took that illustration and tried to get well the unique sign. For that to be attainable, the brand new illustration would by some means need to comprise “simply as a lot” info because the wave we began from. That “simply as a lot” is obtained from the Fourier Rework, and it consists of the magnitudes and section shifts of the completely different frequencies that make up the sign.

How, then, does the Fourier-transformed model of the “chicken” sound wave look? We receive it by calling torch_fft_fft() (the place fft stands for Quick Fourier Rework):

dft <- torch_fft_fft(pattern$waveform)
dim(dft)

[1]     1 16000

The size of this tensor is similar; nevertheless, its values usually are not in chronological order. As a substitute, they symbolize the Fourier coefficients, comparable to the frequencies contained within the sign. The upper their magnitude, the extra they contribute to the sign:

magazine <- torch_abs(dft[1, ])

df <- information.body(
  x = 1:(size(pattern$waveform[1]) / 2),
  y = as.numeric(magazine[1:8000])
)

ggplot(df, aes(x = x, y = y)) +
  geom_line(measurement = 0.3) +
  ggtitle(
    paste0(
      "The spoken phrase "",
      pattern$label,
      "": Discrete Fourier Rework"
    )
  ) +
  xlab("frequency") +
  ylab("magnitude") +
  theme_minimal()

From this alternate illustration, we may return to the unique sound wave by taking the frequencies current within the sign, weighting them in response to their coefficients, and including them up. However in sound classification, timing info should certainly matter; we don’t actually wish to throw it away.

Combining representations: The spectrogram

In actual fact, what actually would assist us is a synthesis of each representations; some kind of “have your cake and eat it, too.” What if we may divide the sign into small chunks, and run the Fourier Rework on every of them? As you’ll have guessed from this lead-up, this certainly is one thing we will do; and the illustration it creates is named the spectrogram.

With a spectrogram, we nonetheless hold some time-domain info – some, since there’s an unavoidable loss in granularity. However, for every of the time segments, we study their spectral composition. There’s an necessary level to be made, although. The resolutions we get in time versus in frequency, respectively, are inversely associated. If we break up up the indicators into many chunks (known as “home windows”), the frequency illustration per window won’t be very fine-grained. Conversely, if we wish to get higher decision within the frequency area, we now have to decide on longer home windows, thus dropping details about how spectral composition varies over time. What appears like a giant drawback – and in lots of circumstances, will likely be – received’t be one for us, although, as you’ll see very quickly.

First, although, let’s create and examine such a spectrogram for our instance sign. Within the following code snippet, the scale of the – overlapping – home windows is chosen in order to permit for cheap granularity in each the time and the frequency area. We’re left with sixty-three home windows, and, for every window, receive 200 fifty-seven coefficients:

fft_size <- 512
window_size <- 512
energy <- 0.5

spectrogram <- transform_spectrogram(
  n_fft = fft_size,
  win_length = window_size,
  normalized = TRUE,
  energy = energy
)

spec <- spectrogram(pattern$waveform)$squeeze()
dim(spec)

[1]   257 63

We are able to show the spectrogram visually:

bins <- 1:dim(spec)[1]
freqs <- bins / (fft_size / 2 + 1) * pattern$sample_rate 
log_freqs <- log10(freqs)

frames <- 1:(dim(spec)[2])
seconds <- (frames / dim(spec)[2]) *
  (dim(pattern$waveform$squeeze())[1] / pattern$sample_rate)

picture(x = as.numeric(seconds),
      y = log_freqs,
      z = t(as.matrix(spec)),
      ylab = 'log frequency [Hz]',
      xlab = 'time [s]',
      col = hcl.colours(12, palette = "viridis")
)
essential <- paste0("Spectrogram, window measurement = ", window_size)
sub <- "Magnitude (sq. root)"
mtext(aspect = 3, line = 2, at = 0, adj = 0, cex = 1.3, essential)
mtext(aspect = 3, line = 1, at = 0, adj = 0, cex = 1, sub)

We all know that we’ve misplaced some decision in each time and frequency. By displaying the sq. root of the coefficients’ magnitudes, although – and thus, enhancing sensitivity – we had been nonetheless capable of receive an inexpensive end result. (With the viridis shade scheme, long-wave shades point out higher-valued coefficients; short-wave ones, the alternative.)

Lastly, let’s get again to the essential query. If this illustration, by necessity, is a compromise – why, then, would we wish to make use of it? That is the place we take the deep-learning perspective. The spectrogram is a two-dimensional illustration: a picture. With photos, we now have entry to a wealthy reservoir of methods and architectures: Amongst all areas deep studying has been profitable in, picture recognition nonetheless stands out. Quickly, you’ll see that for this process, fancy architectures usually are not even wanted; an easy convnet will do an excellent job.

Coaching a neural community on spectrograms

We begin by making a torch::dataset() that, ranging from the unique speechcommand_dataset(), computes a spectrogram for each pattern.

spectrogram_dataset <- dataset(
  inherit = speechcommand_dataset,
  initialize = operate(...,
                        pad_to = 16000,
                        sampling_rate = 16000,
                        n_fft = 512,
                        window_size_seconds = 0.03,
                        window_stride_seconds = 0.01,
                        energy = 2) {
    self$pad_to <- pad_to
    self$window_size_samples <- sampling_rate *
      window_size_seconds
    self$window_stride_samples <- sampling_rate *
      window_stride_seconds
    self$energy <- energy
    self$spectrogram <- transform_spectrogram(
        n_fft = n_fft,
        win_length = self$window_size_samples,
        hop_length = self$window_stride_samples,
        normalized = TRUE,
        energy = self$energy
      )
    tremendous$initialize(...)
  },
  .getitem = operate(i) {
    merchandise <- tremendous$.getitem(i)

    x <- merchandise$waveform
    # ensure that all samples have the identical size (57)
    # shorter ones will likely be padded,
    # longer ones will likely be truncated
    x <- nnf_pad(x, pad = c(0, self$pad_to - dim(x)[2]))
    x <- x %>% self$spectrogram()

    if (is.null(self$energy)) {
      # on this case, there's a further dimension, in place 4,
      # that we wish to seem in entrance
      # (as a second channel)
      x <- x$squeeze()$permute(c(3, 1, 2))
    }

    y <- merchandise$label_index
    listing(x = x, y = y)
  }
)

Within the parameter listing to spectrogram_dataset(), be aware energy, with a default worth of two. That is the worth that, except instructed in any other case, torch’s transform_spectrogram() will assume that energy ought to have. Underneath these circumstances, the values that make up the spectrogram are the squared magnitudes of the Fourier coefficients. Utilizing energy, you possibly can change the default, and specify, for instance, that’d you’d like absolute values (energy = 1), some other constructive worth (akin to 0.5, the one we used above to show a concrete instance) – or each the actual and imaginary components of the coefficients (energy = NULL).

Show-wise, in fact, the total complicated illustration is inconvenient; the spectrogram plot would wish a further dimension. However we could properly ponder whether a neural community may revenue from the extra info contained within the “complete” complicated quantity. In any case, when lowering to magnitudes we lose the section shifts for the person coefficients, which could comprise usable info. In actual fact, my checks confirmed that it did; use of the complicated values resulted in enhanced classification accuracy.

Let’s see what we get from spectrogram_dataset():

ds <- spectrogram_dataset(
  root = "~/.torch-datasets",
  url = "speech_commands_v0.01",
  obtain = TRUE,
  energy = NULL
)

dim(ds[1]$x)

[1]   2 257 101

We’ve got 257 coefficients for 101 home windows; and every coefficient is represented by each its actual and imaginary components.

Subsequent, we break up up the information, and instantiate the dataset() and dataloader() objects.

train_ids <- pattern(
  1:size(ds),
  measurement = 0.6 * size(ds)
)
valid_ids <- pattern(
  setdiff(
    1:size(ds),
    train_ids
  ),
  measurement = 0.2 * size(ds)
)
test_ids <- setdiff(
  1:size(ds),
  union(train_ids, valid_ids)
)

batch_size <- 128

train_ds <- dataset_subset(ds, indices = train_ids)
train_dl <- dataloader(
  train_ds,
  batch_size = batch_size, shuffle = TRUE
)

valid_ds <- dataset_subset(ds, indices = valid_ids)
valid_dl <- dataloader(
  valid_ds,
  batch_size = batch_size
)

test_ds <- dataset_subset(ds, indices = test_ids)
test_dl <- dataloader(test_ds, batch_size = 64)

b <- train_dl %>%
  dataloader_make_iter() %>%
  dataloader_next()

dim(b$x)

[1] 128   2 257 101

The mannequin is an easy convnet, with dropout and batch normalization. The true and imaginary components of the Fourier coefficients are handed to the mannequin’s preliminary nn_conv2d() as two separate channels.

mannequin <- nn_module(
  initialize = operate() {
    self$options <- nn_sequential(
      nn_conv2d(2, 32, kernel_size = 3),
      nn_batch_norm2d(32),
      nn_relu(),
      nn_max_pool2d(kernel_size = 2),
      nn_dropout2d(p = 0.2),
      nn_conv2d(32, 64, kernel_size = 3),
      nn_batch_norm2d(64),
      nn_relu(),
      nn_max_pool2d(kernel_size = 2),
      nn_dropout2d(p = 0.2),
      nn_conv2d(64, 128, kernel_size = 3),
      nn_batch_norm2d(128),
      nn_relu(),
      nn_max_pool2d(kernel_size = 2),
      nn_dropout2d(p = 0.2),
      nn_conv2d(128, 256, kernel_size = 3),
      nn_batch_norm2d(256),
      nn_relu(),
      nn_max_pool2d(kernel_size = 2),
      nn_dropout2d(p = 0.2),
      nn_conv2d(256, 512, kernel_size = 3),
      nn_batch_norm2d(512),
      nn_relu(),
      nn_adaptive_avg_pool2d(c(1, 1)),
      nn_dropout2d(p = 0.2)
    )

    self$classifier <- nn_sequential(
      nn_linear(512, 512),
      nn_batch_norm1d(512),
      nn_relu(),
      nn_dropout(p = 0.5),
      nn_linear(512, 30)
    )
  },
  ahead = operate(x) {
    x <- self$options(x)$squeeze()
    x <- self$classifier(x)
    x
  }
)

We subsequent decide an appropriate studying charge:

mannequin <- mannequin %>%
  setup(
    loss = nn_cross_entropy_loss(),
    optimizer = optim_adam,
    metrics = listing(luz_metric_accuracy())
  )

rates_and_losses <- mannequin %>%
  lr_finder(train_dl)
rates_and_losses %>% plot()

Based mostly on the plot, I made a decision to make use of 0.01 as a maximal studying charge. Coaching went on for forty epochs.

fitted <- mannequin %>%
  match(train_dl,
    epochs = 50, valid_data = valid_dl,
    callbacks = listing(
      luz_callback_early_stopping(endurance = 3),
      luz_callback_lr_scheduler(
        lr_one_cycle,
        max_lr = 1e-2,
        epochs = 50,
        steps_per_epoch = size(train_dl),
        call_on = "on_batch_end"
      ),
      luz_callback_model_checkpoint(path = "models_complex/"),
      luz_callback_csv_logger("logs_complex.csv")
    ),
    verbose = TRUE
  )

plot(fitted)

Let’s verify precise accuracies.

"epoch","set","loss","acc"
1,"practice",3.09768574611813,0.12396992171405
1,"legitimate",2.52993751740923,0.284378862793572
2,"practice",2.26747255972008,0.333642356819118
2,"legitimate",1.66693911248562,0.540791100123609
3,"practice",1.62294889937818,0.518464153275649
3,"legitimate",1.11740599192825,0.704882571075402
...
...
38,"practice",0.18717994078312,0.943809229501442
38,"legitimate",0.23587799138006,0.936418417799753
39,"practice",0.19338578602993,0.942882159044087
39,"legitimate",0.230597475945365,0.939431396786156
40,"practice",0.190593419024368,0.942727647301195
40,"legitimate",0.243536252455384,0.936186650185414

With thirty courses to tell apart between, a closing validation-set accuracy of ~0.94 seems to be like a really first rate end result!

We are able to verify this on the check set:

consider(fitted, test_dl)

loss: 0.2373
acc: 0.9324

An fascinating query is which phrases get confused most frequently. (After all, much more fascinating is how error possibilities are associated to options of the spectrograms – however this, we now have to depart to the true area consultants. A pleasant method of displaying the confusion matrix is to create an alluvial plot. We see the predictions, on the left, “move into” the goal slots. (Goal-prediction pairs much less frequent than a thousandth of check set cardinality are hidden.)

Wrapup

That’s it for at this time! Within the upcoming weeks, anticipate extra posts drawing on content material from the soon-to-appear CRC e book, Deep Studying and Scientific Computing with R torch. Thanks for studying!

Picture by alex lauzon on Unsplash