今日のローカルTensorFlowEverywhereイベントの出欠確認!

SPICE によるピッチ検出

TensorFlow.orgで表示 Google Colab で実行 GitHub でソースを表示 Download notebook TF Hub モデルを見る

この Colab では、TensorFlow Hub からダウンロードした SPICE モデルの使用方法を紹介します。

sudo apt-get install -q -y timidity libsndfile1
Reading package lists...
Building dependency tree...
Reading state information...
libsndfile1 is already the newest version (1.0.28-4ubuntu0.18.04.1).
libsndfile1 set to manually installed.
The following packages were automatically installed and are no longer required:
  adwaita-icon-theme ca-certificates-java dconf-gsettings-backend
  dconf-service default-jre default-jre-headless dkms fonts-dejavu-extra
  freeglut3 freeglut3-dev g++-6 glib-networking glib-networking-common
  glib-networking-services gsettings-desktop-schemas gtk-update-icon-cache
  hicolor-icon-theme humanity-icon-theme java-common libaccinj64-9.1
  libasyncns0 libatk-bridge2.0-0 libatk-wrapper-java libatk-wrapper-java-jni
  libatk1.0-0 libatk1.0-data libatspi2.0-0 libavahi-client3
  libavahi-common-data libavahi-common3 libcairo-gobject2 libcolord2 libcroco3
  libcudart9.1 libcufft9.1 libcufftw9.1 libcups2 libcurand9.1 libcusolver9.1
  libcusparse9.1 libdconf1 libdrm-amdgpu1 libdrm-dev libdrm-intel1
  libdrm-nouveau2 libdrm-radeon1 libegl-mesa0 libegl1 libegl1-mesa libepoxy0
  libfontenc1 libgbm1 libgdk-pixbuf2.0-0 libgdk-pixbuf2.0-common libgif7
  libgl1 libgl1-mesa-dev libgl1-mesa-dri libglapi-mesa libgles1 libgles2
  libglu1-mesa libglu1-mesa-dev libglvnd-core-dev libglvnd-dev libglvnd0
  libglx-mesa0 libglx0 libgtk-3-0 libgtk-3-common libgtk2.0-0 libgtk2.0-common
  libice-dev libjansson4 libjson-glib-1.0-0 libjson-glib-1.0-common liblcms2-2
  libllvm9 libnppc9.1 libnppial9.1 libnppicc9.1 libnppicom9.1 libnppidei9.1
  libnppif9.1 libnppig9.1 libnppim9.1 libnppist9.1 libnppisu9.1 libnppitc9.1
  libnpps9.1 libnvrtc9.1 libnvtoolsext1 libnvvm3 libopengl0 libpciaccess0
  libpcsclite1 libproxy1v5 libpthread-stubs0-dev libpulse0 librest-0.7-0
  librsvg2-2 librsvg2-common libsensors4 libsm-dev libsoup-gnome2.4-1
  libsoup2.4-1 libstdc++-6-dev libthrust-dev libvdpau-dev libvdpau1
  libwayland-client0 libwayland-cursor0 libwayland-egl1 libwayland-server0
  libx11-dev libx11-xcb-dev libx11-xcb1 libxau-dev libxcb-dri2-0
  libxcb-dri2-0-dev libxcb-dri3-0 libxcb-dri3-dev libxcb-glx0 libxcb-glx0-dev
  libxcb-present-dev libxcb-present0 libxcb-randr0 libxcb-randr0-dev
  libxcb-render0-dev libxcb-shape0 libxcb-shape0-dev libxcb-sync-dev
  libxcb-sync1 libxcb-xfixes0 libxcb-xfixes0-dev libxcb1-dev libxcomposite1
  libxcursor1 libxdamage-dev libxdamage1 libxdmcp-dev libxext-dev
  libxfixes-dev libxfixes3 libxfont2 libxft2 libxi-dev libxi6 libxinerama1
  libxkbcommon0 libxkbfile1 libxmu-dev libxmu-headers libxnvctrl0 libxrandr2
  libxshmfence-dev libxshmfence1 libxt-dev libxtst6 libxv1 libxxf86dga1
  libxxf86vm-dev libxxf86vm1 linux-gcp-5.3-headers-5.3.0-1030
  linux-gcp-headers-5.0.0-1026 linux-headers-5.3.0-1030-gcp
  linux-image-5.3.0-1030-gcp linux-modules-5.3.0-1030-gcp
  linux-modules-extra-5.3.0-1030-gcp mesa-common-dev ocl-icd-libopencl1
  ocl-icd-opencl-dev opencl-c-headers openjdk-11-jre openjdk-11-jre-headless
  openjdk-8-jre openjdk-8-jre-headless pkg-config policykit-1-gnome
  python3-xkit screen-resolution-extra ubuntu-mono x11-utils x11-xkb-utils
  x11proto-core-dev x11proto-damage-dev x11proto-dev x11proto-fixes-dev
  x11proto-input-dev x11proto-xext-dev x11proto-xf86vidmode-dev
  xorg-sgml-doctools xserver-common xserver-xorg-core-hwe-18.04 xtrans-dev
Use 'sudo apt autoremove' to remove them.
The following additional packages will be installed:
  freepats libaudio2 libjack-jackd2-0 libsamplerate0 timidity-daemon
Suggested packages:
  nas jackd2 fluid-soundfont-gm fluid-soundfont-gs pmidi
The following NEW packages will be installed:
  freepats libaudio2 libjack-jackd2-0 libsamplerate0 timidity timidity-daemon
0 upgraded, 6 newly installed, 0 to remove and 100 not upgraded.
Need to get 30.8 MB of archives.
After this operation, 38.3 MB of additional disk space will be used.
Get:1 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/universe amd64 freepats all 20060219-1 [29.0 MB]
Get:2 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libaudio2 amd64 1.9.4-6 [50.3 kB]
Get:3 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libsamplerate0 amd64 0.1.9-1 [938 kB]
Get:4 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/main amd64 libjack-jackd2-0 amd64 1.9.12~dfsg-2 [263 kB]
Get:5 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/universe amd64 timidity amd64 2.13.2-41 [585 kB]
Get:6 http://asia-east1.gce.archive.ubuntu.com/ubuntu bionic/universe amd64 timidity-daemon all 2.13.2-41 [5984 B]
Fetched 30.8 MB in 3s (9878 kB/s)
Selecting previously unselected package freepats.
(Reading database ... 242335 files and directories currently installed.)
Preparing to unpack .../0-freepats_20060219-1_all.deb ...
Unpacking freepats (20060219-1) ...
Selecting previously unselected package libaudio2:amd64.
Preparing to unpack .../1-libaudio2_1.9.4-6_amd64.deb ...
Unpacking libaudio2:amd64 (1.9.4-6) ...
Selecting previously unselected package libsamplerate0:amd64.
Preparing to unpack .../2-libsamplerate0_0.1.9-1_amd64.deb ...
Unpacking libsamplerate0:amd64 (0.1.9-1) ...
Selecting previously unselected package libjack-jackd2-0:amd64.
Preparing to unpack .../3-libjack-jackd2-0_1.9.12~dfsg-2_amd64.deb ...
Unpacking libjack-jackd2-0:amd64 (1.9.12~dfsg-2) ...
Selecting previously unselected package timidity.
Preparing to unpack .../4-timidity_2.13.2-41_amd64.deb ...
Unpacking timidity (2.13.2-41) ...
Selecting previously unselected package timidity-daemon.
Preparing to unpack .../5-timidity-daemon_2.13.2-41_all.deb ...
Unpacking timidity-daemon (2.13.2-41) ...
Setting up libsamplerate0:amd64 (0.1.9-1) ...
Setting up freepats (20060219-1) ...
Setting up libaudio2:amd64 (1.9.4-6) ...
Setting up libjack-jackd2-0:amd64 (1.9.12~dfsg-2) ...
Setting up timidity (2.13.2-41) ...
Setting up timidity-daemon (2.13.2-41) ...
Adding group timidity....done
Adding system user timidity....done
Adding user `timidity' to group `audio' ...
Adding user timidity to group audio
Done.
Processing triggers for man-db (2.8.3-2ubuntu0.1) ...
Processing triggers for ureadahead (0.100.0-21) ...
Processing triggers for libc-bin (2.27-3ubuntu1.2) ...
Processing triggers for systemd (237-3ubuntu10.38) ...

# All the imports to deal with sound data
pip install -q pydub numba==0.48 librosa music21
import tensorflow as tf
import tensorflow_hub as hub

import numpy as np
import matplotlib.pyplot as plt
import librosa
from librosa import display as librosadisplay

import logging
import math
import statistics
import sys

from IPython.display import Audio, Javascript
from scipy.io import wavfile

from base64 import b64decode

import music21
from pydub import AudioSegment

logger = logging.getLogger()
logger.setLevel(logging.ERROR)

print("tensorflow: %s" % tf.__version__)
#print("librosa: %s" % librosa.__version__)
/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/numba/errors.py:137: UserWarning: Insufficiently recent colorama version found. Numba requires colorama >= 0.3.9
  warnings.warn(msg)

tensorflow: 2.4.1

/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/pydub/utils.py:170: RuntimeWarning: Couldn't find ffmpeg or avconv - defaulting to ffmpeg, but may not work
  warn("Couldn't find ffmpeg or avconv - defaulting to ffmpeg, but may not work", RuntimeWarning)

音声入力ファイル

これが最も困難な部分です。あなたの歌声を録音しましょう!:)

音声ファイルの取得には、次の 4 つの方法があります。

  1. Colab で直接音声を録音する
  2. ご利用の PC からアップロードする
  3. Google Drive に保存されたファイルを使用する
  4. ウェブからファイルをダウンロードする

以下の 4 つの方法から 1 つを選択してください。

[Run this] Definition of the JS code to record audio straight from the browser

Select how to input your audio

INPUT_SOURCE = 'https://storage.googleapis.com/download.tensorflow.org/data/c-scale-metronome.wav'

print('You selected', INPUT_SOURCE)

if INPUT_SOURCE == 'RECORD':
  uploaded_file_name = record(5)
elif INPUT_SOURCE == 'UPLOAD':
  try:
    from google.colab import files
  except ImportError:
    print("ImportError: files from google.colab seems to not be available")
  else:
    uploaded = files.upload()
    for fn in uploaded.keys():
      print('User uploaded file "{name}" with length {length} bytes'.format(
          name=fn, length=len(uploaded[fn])))
    uploaded_file_name = next(iter(uploaded))
    print('Uploaded file: ' + uploaded_file_name)
elif INPUT_SOURCE.startswith('./drive/'):
  try:
    from google.colab import drive
  except ImportError:
    print("ImportError: files from google.colab seems to not be available")
  else:
    drive.mount('/content/drive')
    # don't forget to change the name of the file you
    # will you here!
    gdrive_audio_file = 'YOUR_MUSIC_FILE.wav'
    uploaded_file_name = INPUT_SOURCE
elif INPUT_SOURCE.startswith('http'):
  !wget --no-check-certificate 'https://storage.googleapis.com/download.tensorflow.org/data/c-scale-metronome.wav' -O c-scale.wav
  uploaded_file_name = 'c-scale.wav'
else:
  print('Unrecognized input format!')
  print('Please select "RECORD", "UPLOAD", or specify a file hosted on Google Drive or a file from the web to download file to download')
You selected https://storage.googleapis.com/download.tensorflow.org/data/c-scale-metronome.wav
--2021-02-12 21:17:58--  https://storage.googleapis.com/download.tensorflow.org/data/c-scale-metronome.wav
Resolving storage.googleapis.com (storage.googleapis.com)... 108.177.125.128, 108.177.97.128, 64.233.189.128, ...
Connecting to storage.googleapis.com (storage.googleapis.com)|108.177.125.128|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 384728 (376K) [audio/wav]
Saving to: ‘c-scale.wav’

c-scale.wav         100%[===================>] 375.71K  --.-KB/s    in 0.003s  

2021-02-12 21:17:58 (106 MB/s) - ‘c-scale.wav’ saved [384728/384728]


音声データを準備する

音声データを入手したので、期待される形式に変換して聴いてみましょう!

SPICE モデルでは、入力として、サンプリングレート 16 kHz の音声ファイルが必要です。また、チャンネル数は 1 つ(Mono)である必要があります。

この部分の作業を支援するために、wav ファイルをモデルが期待する形式に変換する関数(convert_audio_for_model)を用意しました。

# Function that converts the user-created audio to the format that the model 
# expects: bitrate 16kHz and only one channel (mono).

EXPECTED_SAMPLE_RATE = 16000

def convert_audio_for_model(user_file, output_file='converted_audio_file.wav'):
  audio = AudioSegment.from_file(user_file)
  audio = audio.set_frame_rate(EXPECTED_SAMPLE_RATE).set_channels(1)
  audio.export(output_file, format="wav")
  return output_file
# Converting to the expected format for the model
# in all the input 4 input method before, the uploaded file name is at
# the variable uploaded_file_name
converted_audio_file = convert_audio_for_model(uploaded_file_name)
# Loading audio samples from the wav file:
sample_rate, audio_samples = wavfile.read(converted_audio_file, 'rb')

# Show some basic information about the audio.
duration = len(audio_samples)/sample_rate
print(f'Sample rate: {sample_rate} Hz')
print(f'Total duration: {duration:.2f}s')
print(f'Size of the input: {len(audio_samples)}')

# Let's listen to the wav file.
Audio(audio_samples, rate=sample_rate)
Sample rate: 16000 Hz
Total duration: 11.89s
Size of the input: 190316

まずはじめに、歌声の波形を見てみましょう。

# We can visualize the audio as a waveform.
_ = plt.plot(audio_samples)

png

より詳しいビジュアライゼーションとして、経時的に周波数を示すスペクトログラム があります。

ここでは、対数による周波数スケールを使用して、歌声をより鮮明に視覚化させます。

MAX_ABS_INT16 = 32768.0

def plot_stft(x, sample_rate, show_black_and_white=False):
  x_stft = np.abs(librosa.stft(x, n_fft=2048))
  fig, ax = plt.subplots()
  fig.set_size_inches(20, 10)
  x_stft_db = librosa.amplitude_to_db(x_stft, ref=np.max)
  if(show_black_and_white):
    librosadisplay.specshow(data=x_stft_db, y_axis='log', 
                             sr=sample_rate, cmap='gray_r')
  else:
    librosadisplay.specshow(data=x_stft_db, y_axis='log', sr=sample_rate)

  plt.colorbar(format='%+2.0f dB')

plot_stft(audio_samples / MAX_ABS_INT16 , sample_rate=EXPECTED_SAMPLE_RATE)
plt.show()
/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/librosa/display.py:974: MatplotlibDeprecationWarning: The 'basey' parameter of __init__() has been renamed 'base' since Matplotlib 3.3; support for the old name will be dropped two minor releases later.
  scaler(mode, **kwargs)
/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/librosa/display.py:974: MatplotlibDeprecationWarning: The 'linthreshy' parameter of __init__() has been renamed 'linthresh' since Matplotlib 3.3; support for the old name will be dropped two minor releases later.
  scaler(mode, **kwargs)
/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/librosa/display.py:974: MatplotlibDeprecationWarning: The 'linscaley' parameter of __init__() has been renamed 'linscale' since Matplotlib 3.3; support for the old name will be dropped two minor releases later.
  scaler(mode, **kwargs)

png

ここで、最後の変換を行う必要があります。音声サンプルは int16 形式です。これらを -1 と 1 の間の浮動小数点数に正規化する必要があります。

audio_samples = audio_samples / float(MAX_ABS_INT16)

モデルを実行する

ようやく簡単な作業です。TensorFlow Hub でモデルを読み込み、音声をフィードしましょう。SPICE から ピッチと不確実性の 2 つの出力が得られます。

TensorFlow Hub は、機械学習モデルの再利用可能な部分の公開、発見、および消費のためのライブラリです。ユーザーの抱える課題の解決する機械学習の使用を簡単にすることができます。

モデルを読み込むには、Hub モジュールと、モデルにポイントする URL のみが必要です。

# Loading the SPICE model is easy:
model = hub.load("https://tfhub.dev/google/spice/2")
WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'global_step:0' shape=() dtype=int64_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'global_step:0' shape=() dtype=int64_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/conv2d/kernel:0' shape=(1, 3, 1, 64) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/conv2d/kernel:0' shape=(1, 3, 1, 64) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/gamma:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/gamma:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/beta:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/beta:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/moving_mean:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/moving_mean:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'global_step:0' shape=() dtype=int64_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'global_step:0' shape=() dtype=int64_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/conv2d/kernel:0' shape=(1, 3, 1, 64) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/conv2d/kernel:0' shape=(1, 3, 1, 64) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/gamma:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/gamma:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/beta:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/beta:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/moving_mean:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/moving_mean:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'global_step:0' shape=() dtype=int64_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'global_step:0' shape=() dtype=int64_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/conv2d/kernel:0' shape=(1, 3, 1, 64) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/conv2d/kernel:0' shape=(1, 3, 1, 64) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/gamma:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/gamma:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/beta:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/beta:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/moving_mean:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/moving_mean:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'global_step:0' shape=() dtype=int64_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'global_step:0' shape=() dtype=int64_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/conv2d/kernel:0' shape=(1, 3, 1, 64) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/conv2d/kernel:0' shape=(1, 3, 1, 64) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/gamma:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/gamma:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/beta:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/beta:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/moving_mean:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

WARNING:tensorflow:Unable to create a python object for variable <tf.Variable 'encoder/batch_normalization/moving_mean:0' shape=(64,) dtype=float32_ref> because it is a reference variable. It may not be visible to training APIs. If this is a problem, consider rebuilding the SavedModel after running tf.compat.v1.enable_resource_variables().

注意: ここでの豆知識は、Hub のすべてのモデル URL は、ダウンロードだけでなく、ドキュメントの読み取りにも利用できるということです。そのため、ブラウザでそのリンクを開くと、モデルの使用方法を読み、どのようにしてトレーニングされたのかという詳細も知ることができます。

モデルが読み込まれ、データの準備が完了したので、結果を取得するための 3 行を追加しましょう。

# We now feed the audio to the SPICE tf.hub model to obtain pitch and uncertainty outputs as tensors.
model_output = model.signatures["serving_default"](tf.constant(audio_samples, tf.float32))

pitch_outputs = model_output["pitch"]
uncertainty_outputs = model_output["uncertainty"]

# 'Uncertainty' basically means the inverse of confidence.
confidence_outputs = 1.0 - uncertainty_outputs

fig, ax = plt.subplots()
fig.set_size_inches(20, 10)
plt.plot(pitch_outputs, label='pitch')
plt.plot(confidence_outputs, label='confidence')
plt.legend(loc="lower right")
plt.show()

png

信頼度の低い(confidence < 0.9)すべてのピッチ推定値を取り除いて、残りのピッチをグラフ化して、結果を理解しやすくしましょう。

confidence_outputs = list(confidence_outputs)
pitch_outputs = [ float(x) for x in pitch_outputs]

indices = range(len (pitch_outputs))
confident_pitch_outputs = [ (i,p)  
  for i, p, c in zip(indices, pitch_outputs, confidence_outputs) if  c >= 0.9  ]
confident_pitch_outputs_x, confident_pitch_outputs_y = zip(*confident_pitch_outputs)

fig, ax = plt.subplots()
fig.set_size_inches(20, 10)
ax.set_ylim([0, 1])
plt.scatter(confident_pitch_outputs_x, confident_pitch_outputs_y, )
plt.scatter(confident_pitch_outputs_x, confident_pitch_outputs_y, c="r")

plt.show()

png

SPICE が返すピッチ値の範囲は 0 から 1 です。この値を Hz 単位の絶対ピッチ値に変換しましょう。

def output2hz(pitch_output):
  # Constants taken from https://tfhub.dev/google/spice/2
  PT_OFFSET = 25.58
  PT_SLOPE = 63.07
  FMIN = 10.0;
  BINS_PER_OCTAVE = 12.0;
  cqt_bin = pitch_output * PT_SLOPE + PT_OFFSET;
  return FMIN * 2.0 ** (1.0 * cqt_bin / BINS_PER_OCTAVE)

confident_pitch_values_hz = [ output2hz(p) for p in confident_pitch_outputs_y ]

では、どれほど良い予測が得られるか確認しましょう。予測されるピッチを元のスペクトログラムにオーバーレイ表示します。ピッチ予測をより見やすくするために、スペクトログラムをモノクロに変更してます。

plot_stft(audio_samples / MAX_ABS_INT16 , 
          sample_rate=EXPECTED_SAMPLE_RATE, show_black_and_white=True)
# Note: conveniently, since the plot is in log scale, the pitch outputs 
# also get converted to the log scale automatically by matplotlib.
plt.scatter(confident_pitch_outputs_x, confident_pitch_values_hz, c="r")

plt.show()
/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/librosa/display.py:974: MatplotlibDeprecationWarning: The 'basey' parameter of __init__() has been renamed 'base' since Matplotlib 3.3; support for the old name will be dropped two minor releases later.
  scaler(mode, **kwargs)
/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/librosa/display.py:974: MatplotlibDeprecationWarning: The 'linthreshy' parameter of __init__() has been renamed 'linthresh' since Matplotlib 3.3; support for the old name will be dropped two minor releases later.
  scaler(mode, **kwargs)
/tmpfs/src/tf_docs_env/lib/python3.6/site-packages/librosa/display.py:974: MatplotlibDeprecationWarning: The 'linscaley' parameter of __init__() has been renamed 'linscale' since Matplotlib 3.3; support for the old name will be dropped two minor releases later.
  scaler(mode, **kwargs)

png

音符に変換する

ピッチ値を得たので、次はこの値を音符に変換することにしましょう!この作業はそれ自体が困難となる部分です。次の 2 つの項目を考慮する必要があります。

  1. 休符(歌声がないところ)
  2. 各音符のサイズ(オフセット)

1: 出力にゼロを追加して、歌声がない部分を示す

pitch_outputs_and_rests = [
    output2hz(p) if c >= 0.9 else 0
    for i, p, c in zip(indices, pitch_outputs, confidence_outputs)
]

2: 音符のオフセットを追加する

自由に歌う場合、メロディーには音符が表現する絶対ピッチ値に対するオフセットがある場合があります。したがって、予測を音符に変換するには、この潜在的なオフセットを修正する必要があります。次のコードはこれを計算しています。

A4 = 440
C0 = A4 * pow(2, -4.75)
note_names = ["C", "C#", "D", "D#", "E", "F", "F#", "G", "G#", "A", "A#", "B"]

def hz2offset(freq):
  # This measures the quantization error for a single note.
  if freq == 0:  # Rests always have zero error.
    return None
  # Quantized note.
  h = round(12 * math.log2(freq / C0))
  return 12 * math.log2(freq / C0) - h


# The ideal offset is the mean quantization error for all the notes
# (excluding rests):
offsets = [hz2offset(p) for p in pitch_outputs_and_rests if p != 0]
print("offsets: ", offsets)

ideal_offset = statistics.mean(offsets)
print("ideal offset: ", ideal_offset)
offsets:  [0.2851056911175931, 0.3700368844097355, 0.2861620445674262, 0.19609005646164235, 0.17851549283916768, 0.2733467103665532, -0.4475316266590852, -0.24651997073237908, -0.17965956403558891, -0.23060136331860548, -0.3782634107643901, -0.4725100625926686, -0.3457194541269999, -0.2436666886383776, -0.1818906877810207, -0.1348077739650435, -0.24551812662426897, -0.4454884661609313, -0.31267739488426827, -0.12241723670307181, -0.06614479972665066, -0.06702634735648871, -0.1744135098034576, -0.29365551425759406, -0.32520890458170726, -0.0564402572685907, 0.1470525135224534, 0.17167006002122775, 0.16529246704037348, 0.09569531546290477, -0.006323616641203955, -0.11799822075907684, -0.18835098459069144, -0.17934754504506145, -0.17215419157092526, -0.23695640070980062, -0.34594501002376177, -0.39380421205108007, -0.2528674895936689, -0.11009436621014146, -0.07118785365169344, -0.08042248799149121, -0.12799786551538972, -0.16227484329287023, -0.059321733849699854, 0.1066761283701112, 0.2104449983065848, 0.2931939382975841, -0.22329278631751492, -0.12365553720538003, -0.4571117360765271, -0.34864566459005175, -0.35947798653189267, -0.4313175396496476, -0.4818984494978622, 0.44220950977261, 0.45883109973128455, -0.47095522924010425, -0.3674495078498552, -0.3047186536962201, -0.31076167209689487, -0.4501382996017185, 0.3966096259778311, 0.4238116671269694, 0.4982695482795947, -0.45932030423227843, -0.4890504510576079, 0.3836890323584754, 0.4441304941600137, -0.38787359430138935, -0.24855899466817277, -0.20666198684519088, -0.23811575664822726, -0.2760223047310504, -0.3641714288169524, -0.41670715643708434, -0.41009272976462086, -0.3340427999073796, -0.26122771753614416, -0.2232610212141708, -0.19940660549943345, -0.22529102428501346, -0.2780880208188776, -0.2744434134537457, -0.25655307157580154, -0.33068201704567457, -0.4678933079416083, -0.4695135511333177, -0.1648153518015647, -0.2461865211898413, -0.48052406086269883, -0.3771743489677135, -0.32261613680665846, -0.25560160024707557, -0.24629929913823645, -0.14035005553309787, -0.16659160448853783, -0.2442749349648139, -0.236978201704666, -0.20882882578912643, -0.22637331529204374, -0.29836323900762807, -0.39081296219175243, -0.3909896476442043, -0.3650093676025108, -0.2642347521955202, -0.13023199393098395, -0.18214744283501716, -0.3020867909366345, -0.33754229827467697, -0.34391801162306024, -0.31454499496763333, -0.26713502510135356, -0.2910458297902849, -0.11686573876684037, -0.1673094354445226, -0.24345710619036964, -0.30852622314040445, -0.35647376789395935, -0.37154654069487236, -0.3600187547380216, -0.2667062802488047, -0.21902188404148148, -0.2484456507736752, -0.2774107871825038, -0.2941432754570741, -0.31118778272216474, -0.32662896348779213, -0.3053947554403962, -0.2160201109821145, -0.17343703730647775, -0.17792559965198507, -0.19880455716197787, -0.2725011871630372, -0.3152120758468442, -0.28217189623658356, -0.11595223738495974, 0.0541902144377957, 0.11488166735824024, -0.2559698195630773, 0.019304235739056708, -0.002236352401425279, 0.44688340799267223, 0.15514959977323883, 0.4207713650291751, 0.3854436726992816, 0.4373497234409598, -0.4694994504625001, -0.3662719146782649, -0.20353897406404542, -0.015043790774988963, -0.4185651697093675, -0.17896653874461066, -0.032896162706066434, -0.061098168330843805, -0.1953772325689087, -0.2545161090666568, -0.3363741032654488, -0.39191536320988973, -0.36531668408458984, -0.3489657612020167, -0.35455202891175475, -0.38925380362813655, 0.48781635300571935, -0.2820884378129733, -0.241939488189864, -0.24987341685836384, -0.3034880535179809, -0.2910712014014081, -0.2783103765422581, -0.30017802073304267, -0.23735882385318519, -0.15802705569807785, -0.1688725350672513, 0.00533368216211727, -0.2545762573057857, -0.28210347487274845, -0.29791494323556833, -0.3228369901949648, -0.3895802937323367, 0.4323827980583488, 0.17438820408041522, -0.12961415393892395, -0.2236258517081069, -0.04022635205333813, -0.4264043621594098, -0.0019025255615048309, -0.07466309859101727, -0.08665327413623203, -0.08169104440753472, -0.31617519541327965, -0.47420548422877573, 0.1502044753855003, 0.30507923857624064, 0.031032583278971515, -0.17852388186996393, -0.3371385477358615, -0.41780861421172233, -0.2023970939094255, -0.10605277224127718, -0.10770872844999246, -0.16037790997569346, -0.18698410763089868, -0.17355977250879562, -0.008242337244190878, -0.011401999431292609, -0.1876701734835322, -0.360175323324853, 0.011681766969516616, -0.1931417836124183]
ideal offset:  -0.16889353086013528

ヒューリスティックを使用して歌われた可能性の最も高い音符のシーケンスの予測を試みることができるようになりました。上記で計算された理想的なオフセットは 1 つの材料ではありますが、速度(8 個など、どれくらいの予測をたてるのか)と量子化を始める時間オフセットを知る必要もあります。単純にしておくために、異なる速度と時間オフセットを試して、量子化誤差を測定し、最終的に、この誤差を最小限に抑える値を使用します。

def quantize_predictions(group, ideal_offset):
  # Group values are either 0, or a pitch in Hz.
  non_zero_values = [v for v in group if v != 0]
  zero_values_count = len(group) - len(non_zero_values)

  # Create a rest if 80% is silent, otherwise create a note.
  if zero_values_count > 0.8 * len(group):
    # Interpret as a rest. Count each dropped note as an error, weighted a bit
    # worse than a badly sung note (which would 'cost' 0.5).
    return 0.51 * len(non_zero_values), "Rest"
  else:
    # Interpret as note, estimating as mean of non-rest predictions.
    h = round(
        statistics.mean([
            12 * math.log2(freq / C0) - ideal_offset for freq in non_zero_values
        ]))
    octave = h // 12
    n = h % 12
    note = note_names[n] + str(octave)
    # Quantization error is the total difference from the quantized note.
    error = sum([
        abs(12 * math.log2(freq / C0) - ideal_offset - h)
        for freq in non_zero_values
    ])
    return error, note


def get_quantization_and_error(pitch_outputs_and_rests, predictions_per_eighth,
                               prediction_start_offset, ideal_offset):
  # Apply the start offset - we can just add the offset as rests.
  pitch_outputs_and_rests = [0] * prediction_start_offset + \
                            pitch_outputs_and_rests
  # Collect the predictions for each note (or rest).
  groups = [
      pitch_outputs_and_rests[i:i + predictions_per_eighth]
      for i in range(0, len(pitch_outputs_and_rests), predictions_per_eighth)
  ]

  quantization_error = 0

  notes_and_rests = []
  for group in groups:
    error, note_or_rest = quantize_predictions(group, ideal_offset)
    quantization_error += error
    notes_and_rests.append(note_or_rest)

  return quantization_error, notes_and_rests


best_error = float("inf")
best_notes_and_rests = None
best_predictions_per_note = None

for predictions_per_note in range(20, 65, 1):
  for prediction_start_offset in range(predictions_per_note):

    error, notes_and_rests = get_quantization_and_error(
        pitch_outputs_and_rests, predictions_per_note,
        prediction_start_offset, ideal_offset)

    if error < best_error:      
      best_error = error
      best_notes_and_rests = notes_and_rests
      best_predictions_per_note = predictions_per_note

# At this point, best_notes_and_rests contains the best quantization.
# Since we don't need to have rests at the beginning, let's remove these:
while best_notes_and_rests[0] == 'Rest':
  best_notes_and_rests = best_notes_and_rests[1:]
# Also remove silence at the end.
while best_notes_and_rests[-1] == 'Rest':
  best_notes_and_rests = best_notes_and_rests[:-1]

では、量子化された音符を楽譜として書き出しましょう!

これを行うには、music21Open Sheet Music Display の 2 つのライブラリを使用します。

注意: 単純に行えるように、ここではすべての音符の長さが同じ(半音)であると仮定しています。

# Creating the sheet music score.
sc = music21.stream.Score()
# Adjust the speed to match the actual singing.
bpm = 60 * 60 / best_predictions_per_note
print ('bpm: ', bpm)
a = music21.tempo.MetronomeMark(number=bpm)
sc.insert(0,a)

for snote in best_notes_and_rests:   
    d = 'half'
    if snote == 'Rest':      
      sc.append(music21.note.Rest(type=d))
    else:
      sc.append(music21.note.Note(snote, type=d))
bpm:  78.26086956521739

[Run this] Helper function to use Open Sheet Music Display (JS code) to show a music score

from IPython.core.display import display, HTML, Javascript
import json, random

def showScore(score):
    xml = open(score.write('musicxml')).read()
    showMusicXML(xml)

def showMusicXML(xml):
    DIV_ID = "OSMD_div"
    display(HTML('<div id="'+DIV_ID+'">loading OpenSheetMusicDisplay</div>'))
    script = """
    var div_id = { {DIV_ID} };
    function loadOSMD() { 
        return new Promise(function(resolve, reject){
            if (window.opensheetmusicdisplay) {
                return resolve(window.opensheetmusicdisplay)
            }
            // OSMD script has a 'define' call which conflicts with requirejs
            var _define = window.define // save the define object 
            window.define = undefined // now the loaded script will ignore requirejs
            var s = document.createElement( 'script' );
            s.setAttribute( 'src', "https://cdn.jsdelivr.net/npm/opensheetmusicdisplay@0.7.6/build/opensheetmusicdisplay.min.js" );
            //s.setAttribute( 'src', "/custom/opensheetmusicdisplay.js" );
            s.onload=function(){
                window.define = _define
                resolve(opensheetmusicdisplay);
            };
            document.body.appendChild( s ); // browser will try to load the new script tag
        }) 
    }
    loadOSMD().then((OSMD)=>{
        window.openSheetMusicDisplay = new OSMD.OpenSheetMusicDisplay(div_id, {
          drawingParameters: "compacttight"
        });
        openSheetMusicDisplay
            .load({ {data} })
            .then(
              function() {
                openSheetMusicDisplay.render();
              }
            );
    })
    """.replace('{ {DIV_ID} }',DIV_ID).replace('{ {data} }',json.dumps(xml))
    display(Javascript(script))
    return
# rendering the music score
showScore(sc)
print(best_notes_and_rests)
<IPython.core.display.Javascript object>
['C3', 'D3', 'E3', 'F3', 'G3', 'A3', 'B3', 'C4']

音符を MIDI ファイルに変換して、聴いてみましょう。

このファイルを作成するには、前に作成したストリームを使用できます。

# Saving the recognized musical notes as a MIDI file
converted_audio_file_as_midi = converted_audio_file[:-4] + '.mid'
fp = sc.write('midi', fp=converted_audio_file_as_midi)
wav_from_created_midi = converted_audio_file_as_midi.replace(' ', '_') + "_midioutput.wav"
print(wav_from_created_midi)
converted_audio_file.mid_midioutput.wav

Colab で聴くには、wav に変換し直す必要があります。簡単な方法は、Timidty を使用することです。

timidity $converted_audio_file_as_midi -Ow -o $wav_from_created_midi
Playing converted_audio_file.mid
MIDI file: converted_audio_file.mid
Format: 1  Tracks: 2  Divisions: 1024
Track name: 
Playing time: ~16 seconds
Notes cut: 0
Notes lost totally: 0

最後に、モデルが推論し、予測されたピッチから MIDI 経由で作成され、音符から作成された音声を聴いてみましょう!

Audio(wav_from_created_midi)