Zero-based realization of singing voice recognition and classification system)
Chapter One Introduction
This article describes the process of how a deep learning Xiaobai develops a voice recognition and evaluation system, starting from zero. I will put the Baidu cloud link on the relevant learning materials. If the acquisition expires or fails, leave a message in the comment area, and I Will reissue. This is the blogger's first article, please forgive me for the bad writing!
clear purpose
The purpose of this experiment is to classify the high, middle, and low pitches of singing voices (male and female), so it is necessary to understand the so-called differences between audio.
Human beings control the throat muscles and pull the vocal cords to change the tension and thickness, and let the air flow impact the vocal cords. When the vocal cords vibrate, the changes in frequency can produce the sounds of the three vocal regions of high, middle and low. The high voice area is also called the head voice area, the middle voice area is also called the mixed voice area, and the low voice area is also called the chest voice area. The concept of the voice area is different from the main voice in art singing. The six classifications of tenor high school bass and female high school bass just represent different ranges, and in music theory, each range has three separate sound zones.
Classified in the broad sense of vocal music, the soprano's range is c1-a2, a total of 15 degrees. According to the characteristics of timbre and range, it can be divided into: coloratura soprano, lyric soprano, drama soprano, lyric coloratura soprano and lyric drama soprano Treble [1]; mezzo-soprano range a-a2, the higher mezzo-soprano is similar to the dramatic soprano, and similarly, the low mezzo-soprano is very close to the alto, which is nothing to those who have no musical foundation The difference; the alto range is f-f2, and both male and female basses are easy to distinguish. The sound is relatively thick and solid, as if there is a foreign object pressing down on the throat; the tenor range is c1-a2, which can also be divided into lyrical tenor, Dramatic tenor and light tenor have a range of c1-a2; baritones are a-f2; basses have a range of g-d2. The respective categories are shown in Table 1.
Environment installation
We first install the Python environment. It is recommended that the installation of anaconda and TensorFlow should be downloaded from the official website in the future.
The Python version number is 3.6.3 (https://www.python.org/downloads/release/python-363/). I used 3.7 or above before, but it was not easy to use later, so I changed it to 3.6.
After downloading and installing , but also to configure the environment variables, please Baidu for details, and I won’t introduce it in detail here. After success, the picture is as follows:
The following is the installation of anaconda, here is a direct link to Baidu Cloud:
Link: https://pan.baidu.com/s/1_aFCyFTRiPrxlIO7rxpv_w
Extraction code: jhy4
What you need to remember is that you need to install it to a certain step Check the environment variable (red is normal):
Then start the installation of TensorFlow.
The blogger’s computer is i5-4210MQ\GPU is GTX965M. In contrast, the GPU version must be installed. Of course, I will also mention the CPU version
1. CPU version:
suitable for integrated graphics cards, all amd graphics cards, or computers lower than Nvidia gt750M
Enter in cmd:
pip install tensorflow==2.0.0 -i https://pypi.douban.com/simple --user (==2.0.0 indicates the installed version --user indicates permissions, -i https: //pypi.douban.com/simple indicates that downloading the image from this link can speed up the installation.)
2. Install tensorflow with GPU (the computer must have an Nvidia graphics card, and the performance must be above gt750M level----if your computer has only been purchased for one year, it will definitely work, and if it is an Nvidia graphics card, it will definitely work) Download cuda9.0
and cudnn 7.0.5
open
https://developer.nvidia.com/cuda-90-download-archive
and download as follows (take win10 as an example):
Cudnn is downloaded as follows:
Open https://developer.nvidia.com/rdp/cudnn-download
After both Cuda9.0 and cudnn7.0.5 are downloaded, start to install cuda9.0.
Double-click the cuda9.0 download file, and then follow the default selection until the installation option, check as follows:
Then click Next, unzip Cudnn7.0.5 to get a cuda folder, copy all the files in it, and paste it to the following location: Then
enter the installation in cmd:
pip install tensorflow-gpu2.0.0 -i https://pypi.douban.com/simple --The
library that the user will use later, such as keras, can be directly entered in the windows terminal:
pip install keras2.2.5
If no error is reported, it means success!
At present, the entire environment has been installed, and we can proceed to the next step of learning