• Stars
    star
    106
  • Rank 313,320 (Top 7 %)
  • Language
    Python
  • License
    Apache License 2.0
  • Created over 1 year ago
  • Updated 2 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Easier audio-based machine learning with TensorFlow.

License PyPI - Python Version Supported Platforms Lifecycle

realbook 📒

Realbook is a Python library for easier training of audio deep learning models with Tensorflow made by Spotify's Spotify's Audio Intelligence Lab. Realbook provides callbacks (e.g., spectrogram visualization) and well-tested Keras layers (e.g., STFT, ISTFT, magnitude spectrogram) that we often use when training. These functions have helped standardized consistency across all of our models we and hope realbook will do the same for the open source community.

Notable Features

Below are a few highlights of what we have written so far.

Keras Layers

  • FrozenGraphLayer - Allows you to use a TF V1 graph as a Keras layer.
  • CQT - Constant-Q transform layers ported from nnAudio.
  • Stft, Istft, MelSpectrogram, Spectrogram, Magnitude, Phase and MagnitudeToDecibel - Layers that perform common audio feature preprocessing. All checked for correctness against librosa.

Callbacks

  • Spectrogram visualization - Allows you to write spectrogram output layers to TensorBoard.
  • Training Speed - Allows you to visualize on TensorBoard how fast each epoch of training is taking.
  • Utilization - Allows you to plot on TensorBoard CPU, CPU Memory, GPU and GPU Memory utilization as you train.

Installation

pip install realbook

# Or, if using any TensorBoard-related callbacks, install additional dependencies:
pip install realbook[tensorboard]

Then, in your code:

import realbook.callbacks.spectrogram_visualization # a nifty TensorBoard callback

Example

A Binary Classifier With Audio Input

Let's use realbook to train a binary classifier that takes in audio, converts the audio to a spectrogram and then runs the spectorgram output through two trainable Dense layers.

import tensorflow as tf
from realbook.layers.signal import STFT

train_ds = tf.data.TFRecordDataset(training_filenames)
val_ds = tf.data.TFRecordDataset(validation_filenames)

# Create a sequential model
model = tf.keras.Sequential([
    tf.keras.layers.InputLayer((22050,)),
    signal.Stft(fft_length=1024, hop_length=512), 1_266_384),
    tf.keras.layers.Dense(1024, activation="relu"),
    tf.keras.layers.Dense(2),
])

# Compile the model
model.compile(optimizer='adam',
              loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

# Now train!
model.fit(
  train_ds,
  validation_data=val_ds,
  epochs=epochs
)

A Binary Classifier With Audio Input and CPU Memory Utilization Measurement

Below is the previous binary classifier example, but we're now going to add a realbook callback to the model's callback list.

import tensorflow as tf
from realbook.layers.signal import STFT
from realbook.callbacks.utilization import MemoryUtilizationCallback

train_ds = tf.data.TFRecordDataset(training_filenames)
val_ds = tf.data.TFRecordDataset(validation_filenames)

# Create a sequential model
model = tf.keras.Sequential([
    tf.keras.layers.InputLayer((22050,)),
    signal.Stft(fft_length=1024, hop_length=512), 1_266_384),
    tf.keras.layers.Dense(1024, activation="relu"),
    tf.keras.layers.Dense(2),
])

# Compile the model
model.compile(optimizer='adam',
              loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

writer = tf.summary.create_file_writer(tensorboard_output_location)

# Now train!
model.fit(
  train_ds,
  validation_data=val_ds,
  epochs=epochs,
  callbacks=[MemoryUtilizationCallback(writer))]
)

Metrics

Realbook contains a number of layers that convert audio data (i.e.: waveforms) into various spectral representations (i.e.: spectrograms). For convenience, the amount of memory required for the most commonly used layers is provided below.

Using an FFT length of 1024 and a hop length of 512, processing one second of audio at 22050Hz requires:

Layer Memory High Watermark
realbook.layers.signal.STFT 1,266,384 bytes
realbook.layers.signal.Spectrogram 1,264,324 bytes
realbook.layers.signal.MelSpectrogram 1,262,784 bytes
realbook.layers.nnaudio.CQT 1,047,216 bytes

GPU Utilization Callbacks

GPU resource utilization callbacks are included as part of the tensorboard extra installable. These callbacks expect the program nvidia-smi to be installed. A program which is only available on Linux. For example, on Ubuntu, you can install this program with

apt-get update
apt-get install -y nvidia-utils-<CUDA version number>

Where CUDA version number is the version of CUDA installed on your machine e.g. 450.

Setup Development (of realbook)

Create a new virtual environment with for your supported Python version and clone this repo. Within that virtualenv:

$ pip install -e .[dev]

This will install development dependencies, followed by installing this package itself as "editable".

Run Tests

Tests can be invoked in two ways: pytest and tox.

Run tests via pytest

This must be done within the virtualenv. Note that pytest will automatically pick up the config set in tox.ini. Comment it out if you want to skip coverage and/or ignore verbosity while iterating.

# for all tests
(env) $ pytest tests/

# for one module of tests
(env) $ pytest tests/layers/signal.py

# for one specific test
(env) $ pytest tests/layers/signal.py::test_stft

More info about pytest can be found here.

Run tests via tox

tox should be run outside of a virtualenv. This is because tox will create separate virtual environments for each test environment. A test environment could be based on python versions, or could be specific to documentation, or whatever else. See tox.ini as an example for mulutiple different test environments including: running tests for Python, linting, and checking MANIFEST.in to assert a proper setup.

# run all environments
$ tox

# run a specific environment
$ tox -e check-formatting
$ tox -e py38

Formatting files

Before committing PR's please format your files using tox as some of the formatting options realboook uses is different than the defaults of the Black formatter:

tox -e format

See tox's documentation for more information.

Copyright and License

realbook is Copyright 2022 Spotify AB.

This software is licensed under the Apache License, Version 2.0 (the "Apache License"). You may choose either license to govern your use of this software only upon the condition that you accept all of the terms of either the Apache License.

You may obtain a copy of the Apache License at:

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the Apache License or the GPL License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the Apache License for the specific language governing permissions and limitations under the Apache License

More Repositories

1

luigi

Luigi is a Python module that helps you build complex pipelines of batch jobs. It handles dependency resolution, workflow management, visualization etc. It also comes with Hadoop support built in.
Python
17,089
star
2

annoy

Approximate Nearest Neighbors in C++/Python optimized for memory usage and loading/saving to disk
C++
12,458
star
3

docker-gc

INACTIVE: Docker garbage collection of containers and images
Shell
5,068
star
4

pedalboard

🎛 🔊 A Python library for audio.
C++
4,811
star
5

chartify

Python library that makes it easy for data scientists to create charts.
Python
3,447
star
6

basic-pitch

A lightweight yet powerful audio-to-MIDI converter with pitch bend detection
Python
2,818
star
7

dockerfile-maven

MATURE: A set of Maven tools for dealing with Dockerfiles
Java
2,730
star
8

docker-maven-plugin

INACTIVE: A maven plugin for Docker
Java
2,652
star
9

scio

A Scala API for Apache Beam and Google Cloud Dataflow.
Scala
2,485
star
10

helios

Docker container orchestration platform
Java
2,097
star
11

web-api-examples

Basic examples to authenticate and fetch data using the Spotify Web API
HTML
1,889
star
12

HubFramework

DEPRECATED – Spotify’s component-driven UI framework for iOS
Objective-C
1,864
star
13

apollo

Java libraries for writing composable microservices
Java
1,648
star
14

dh-virtualenv

Python virtualenvs in Debian packages
Python
1,590
star
15

docker-client

INACTIVE: A simple docker client for the JVM
Java
1,425
star
16

docker-kafka

Kafka (and Zookeeper) in Docker
Shell
1,400
star
17

SPTPersistentCache

Everyone tries to implement a cache at some point in their iOS app’s lifecycle, and this is ours.
Objective-C
1,244
star
18

mobius

A functional reactive framework for managing state evolution and side-effects.
Java
1,205
star
19

sparkey

Simple constant key/value storage library, for read-heavy systems with infrequent large bulk inserts.
C
1,143
star
20

ruler

Gradle plugin which helps you analyze the size of your Android apps.
Kotlin
1,100
star
21

voyager

🛰️ Voyager is an approximate nearest-neighbor search library for Python and Java with a focus on ease of use, simplicity, and deployability.
C++
1,090
star
22

XCMetrics

XCMetrics is the easiest way to collect Xcode build metrics and improve developer productivity.
Swift
1,074
star
23

web-api

This issue tracker is no longer used. Join us in the Spotify for Developers forum for support with the Spotify Web API ➡️ https://community.spotify.com/t5/Spotify-for-Developers/bd-p/Spotify_Developer
RAML
981
star
24

echoprint-codegen

Codegen for Echoprint
C++
948
star
25

snakebite

A pure python HDFS client
Python
859
star
26

heroic

The Heroic Time Series Database
Java
843
star
27

klio

Smarter data pipelines for audio.
Python
827
star
28

XCRemoteCache

Swift
815
star
29

apps-tutorial

A Spotify App that contains working examples of the use of Spotify Apps API
627
star
30

SPTDataLoader

The HTTP library used by the Spotify iOS client
Objective-C
624
star
31

ios-sdk

Spotify SDK for iOS
Objective-C
609
star
32

postgresql-metrics

Tool that extracts and provides metrics on your PostgreSQL database
Python
584
star
33

JniHelpers

Tools for writing great JNI code
C++
584
star
34

reactochart

📈 React chart component library 📉
JavaScript
548
star
35

Mobius.swift

A functional reactive framework for managing state evolution and side-effects [Swift implementation]
Swift
540
star
36

dockerfile-mode

An emacs mode for handling Dockerfiles
Emacs Lisp
520
star
37

threaddump-analyzer

A JVM threaddump analyzer
JavaScript
482
star
38

featran

A Scala feature transformation library for data science and machine learning
Scala
467
star
39

android-sdk

Spotify SDK for Android
HTML
440
star
40

echoprint-server

Server for the Echoprint audio fingerprint system
Java
398
star
41

web-scripts

DEPRECATED: A collection of base configs and CLI wrappers used to speed up development @ Spotify.
TypeScript
382
star
42

completable-futures

Utilities for working with futures in Java 8
Java
378
star
43

SpotifyLogin

Swift framework for authenticating with the Spotify API
Swift
344
star
44

ratatool

A tool for data sampling, data generation, and data diffing
Scala
334
star
45

fmt-maven-plugin

Opinionated Maven Plugin that formats your Java code.
Java
299
star
46

big-data-rosetta-code

Code snippets for solving common big data problems in various platforms. Inspired by Rosetta Code
Scala
286
star
47

trickle

A small library for composing asynchronous code
Java
284
star
48

coordinator

A visual interface for turning an SVG into XY coördinates.
HTML
282
star
49

pythonflow

🐍 Dataflow programming for python.
Python
279
star
50

styx

"The path to execution", Styx is a service that schedules batch data processing jobs in Docker containers on Kubernetes.
Java
267
star
51

cstar

Apache Cassandra cluster orchestration tool for the command line
Python
254
star
52

netty-zmtp

A Netty implementation of ZMTP, the ZeroMQ Message Transport Protocol.
Java
242
star
53

ios-style

Guidelines for iOS development in use at Spotify
240
star
54

cassandra-reaper

Software to run automated repairs of cassandra
235
star
55

confidence

Python
232
star
56

spotify-web-api-ts-sdk

A Typescript SDK for the Spotify Web API with types for returned data.
TypeScript
231
star
57

docker-cassandra

Cassandra in Docker with fast startup
Shell
219
star
58

terraform-gke-kubeflow-cluster

Terraform module for creating GKE clusters to run Kubeflow
HCL
209
star
59

dns-java

DNS wrapper library that provides SRV lookup functionality
Java
203
star
60

linux

Spotify's Linux kernel for Debian-based systems
C
203
star
61

git-test

test your commits
Shell
202
star
62

SPStackedNav

[DEPRECATED] Navigation controller which represents its content in stacks of panes, rather than one at a time
Objective-C
195
star
63

basic-pitch-ts

A lightweight yet powerful audio-to-MIDI converter with pitch bend detection.
TypeScript
194
star
64

quickstart

A CommonJS module resolver, loader and compiler for node.js and browsers.
JavaScript
193
star
65

spotify-json

Fast and nice to use C++ JSON library.
C++
190
star
66

dbeam

DBeam exports SQL tables into Avro files using JDBC and Apache Beam
Java
181
star
67

flink-on-k8s-operator

Kubernetes operator for managing the lifecycle of Apache Flink and Beam applications.
Go
178
star
68

bazel-tools

Tools for dealing with very large Bazel-managed repositories
Java
165
star
69

lingon

A user friendly tool for building single-page JavaScript applications
JavaScript
162
star
70

dataenum

Algebraic data types in Java.
Java
159
star
71

magnolify

A collection of Magnolia add-on modules
Scala
157
star
72

async-google-pubsub-client

[SUNSET] Async Google Pubsub Client
Java
156
star
73

gcp-audit

A tool for auditing security properties of GCP projects.
Python
156
star
74

spark-bigquery

Google BigQuery support for Spark, SQL, and DataFrames
Scala
154
star
75

flo

A lightweight workflow definition library
Java
146
star
76

folsom

An asynchronous memcache client for Java
Java
143
star
77

should-up

Remove most of the "should" noise from your tests
JavaScript
143
star
78

missinglink

Build time tool for detecting link problems in java projects
Java
142
star
79

zoltar

Common library for serving TensorFlow, XGBoost and scikit-learn models in production.
Java
141
star
80

android-auth

Spotify authentication and authorization for Android. Part of the Spotify Android SDK.
HTML
139
star
81

proto-registry

An implementation of the Protobuf Registry API
TypeScript
139
star
82

futures-extra

Java library for working with Guava futures
Java
136
star
83

annoy-java

Approximate nearest neighbors in Java
Java
134
star
84

spydra

Ephemeral Hadoop clusters using Google Compute Platform
Java
132
star
85

spotify-tensorflow

Provides Spotify-specific TensorFlow helpers
Python
124
star
86

docker-stress

Simple docker stress test and monitoring tools
Python
124
star
87

spotify-web-playback-sdk-example

React based example app that creates a new player in Spotify Connect to play music from in the browse using Spotify Web Playback SDK.
JavaScript
120
star
88

crtauth

a public key backed client/server authentication system
Python
118
star
89

redux-location-state

Utilities for reading & writing Redux store state to & from the URL
JavaScript
118
star
90

sparkey-java

Java implementation of the Sparkey key value store
Java
117
star
91

rspec-dns

Easily test your DNS with RSpec
Ruby
108
star
92

web-playback-sdk

This issue tracker is no longer used. Join us in the Spotify for Developers forum for support with the Spotify Web Playback SDK ➡️ https://community.spotify.com/t5/Spotify-for-Developers/bd-p/Spotify_Developer
108
star
93

ffwd-ruby

An event and metrics fast-forwarding agent.
Ruby
106
star
94

github-java-client

A Java client to Github API
Java
105
star
95

gimme

Creating time bound IAM Conditions with ease and flair
Python
103
star
96

super-smash-brogp

Sends and withdraws BGP prefixes for fun.
Python
98
star
97

lighthouse-audit-service

TypeScript
93
star
98

noether

Scala Aggregators used for ML Model metrics monitoring
Scala
91
star
99

python-graphwalker

Python re-implementation of the graphwalker testing tool
Python
90
star
100

spotify-js-challenge

JavaScript
87
star