Since the media of this year have started to use artificial intelligence to assist their creation, and artificial intelligence is able to extract the most popular Internet material, and then generate a new article out on this basis. The software is now very mature, more widely used artificial intelligence tool called Get a smart writing in the content business era, many content distribution platform to create new content into the exploratory stage, but its high valuation. Recommendations from algorithm to algorithm to create content "help" the author, the title of today's artificial intelligence has taken forward a big step in the field of content production.
Here we look at the music associated with the creation of artificial intelligence:
Music is one of humanity's natural and artistic hobby to pursue. Since ancient times philosophers of how humans appreciation of the arts, how to create a strong interest. And when early humans would have a lot of music theory. In music history, composers have never stopped to review and summarize the compositional techniques. With the proposed modernization of development, the concept of artificial intelligence, can you let people gradually produced a computer algorithm to use music creation idea, so algorithmic composition concept came into being.
Value
Computer current level or primary composition, the chord portion is realized by manual intervention. Although still in its infancy, although still in its infancy, but the prospect of this technology is quite extensive, display space in several scenes advertising, social, entertainment, AR, VR, etc. are. Imagine, just upload newlyweds after the wedding, you can let the artificial intelligence to customize the exclusive music. In addition, this technology is also used in gaming. Therefore, the use of artificial intelligence to study music composition is necessary.
difficulty
1. knowledge of regular expressions question
Any algorithmic composition system does not exist in a perfect representation techniques melodic development.
2. Innovation music
Automatically by a computer-generated music, composer for the user or whether it is practical? This involves an important issue, that we use algorithmic composition in such a way to mimic the music composer's own music process, or to imitate the style of work of certain types of questions.
3. How to assess the problem of computer-generated musical works
Whether it is generated by an algorithm composing a piece of music can truly meet the traditional music theory? Obviously, quality assessment mechanism composing system is a very important part. It tends to guide the direction of the music, and even ultimately determine the success of musical works.
4. how to give human emotions work for algorithmic composition
reward
1. Read the related algorithmic composition of the material, how to design algorithms have a certain understanding, exercise their ability to find information, useful for future learning.
2. knowledge of the rules of the system, Markov chains, deep learning algorithms have a certain understanding.
3. In the face of problems Why should think more, and more to find the knowledge it contains from little things around, should not complain about their problems, or solve the problem of optimization and improvement should be found from these problems.
4. When you encounter a big difficult problem, can not be intimidated, but should calm down, the big problem refine and identify the issues most essential (ie minimum operational point), then all broken up, Finally, find the problem becomes clear slowly.
Overview
Through access to information, there are many ways algorithm composer currently:
1. knowledge rule system
We have established a series of rules knowledge in the creation of a melody composer collection systems, computer-generated melody requires each produced by a given set of knowledge rules. Rule-based knowledge systems in complex musical creative process, the advantage is very clear knowledge of coding, it can be concluded from a complex relationship. But if you try to imitate all the rules and techniques based composer composer music used to be extracted, entered into the computer, this trend is very difficult.
2. Markov chain
Markov Chain algorithm has been widely used in the field of composition, in accordance with the conversion table to select a note. This conversion table is like a function whose argument is the current notes, and the function value is the possibility to appear next notes. · Conversion table may be set in accordance with standard artefacts, wherein a particular style of music nested. For a particular (such as a particular composer or period of) styles of music into one thousand Guang collection and statistics, can be constructed corresponding conversion table. And this conversion table defines the likelihood of these specific musical style of note-oriented works. In fact, the performance of the conversion table is melodic style quantified. Markov chain mainly for generating a melody of a certain style. This method can simulate thinking when composers create music, to control the computer generates the appropriate music. We can further through the existing melody as an input parameter, or custom music likelihood of each note that appears to produce a new style as output.
3. deep learning
Generating an output related to calculation from an input can be represented by a flow diagram (flow graph): it can represent a flow diagram of FIG calculation, in which FIG each node represents a basic calculation and a calculation value (calculated result value is applied to the node of the child node). Such a collection of computing consideration, it may be allowed at each node and FIG possible construction, and defines a family of functions. Input node without a father, the output node has no children.
Bottom blue part is the music you entered. As can be seen from the figure, it is one of the input points, each input and then incorporated by generating gray point, after the connection layers, the final Output is newly generated music, i.e. ××× point. ××× pulled down to the point, is generated in the new notes. According to a note it is to generate all the music before.
4.SOFM
Self-organizing map (on SOFM) is from one organization network, a so-called self-organization refers to the result of learning is always the weighting vectors of neurons cluster region Maintain the approximation to the input vector so that the input vector having similar characteristics together. This structure can identify the network from input information laws and relationships, and to correspondingly adjust the network in accordance with these rules, so that subsequent output corresponding.
analysis
By reading these data, these methods have a certain understanding, that I myself would happen with artificial intelligence compose it?
For the algorithm to music, my idea is this:
1. heuristic information, search for a melody portion of a length of 10 possible solutions.
Description:
(1) heuristic information: heuristic information on this can be a tone or a few notes, chords and extent of this match on one or a few chords, and this can analyze a large number of existing works, in order to obtain a probability table. For example, if a lot of music, the probability of two chords appear together very small, indicating that the combination of these two chords are discordant, there is no need to continue the search down.
(2) length of 10: Search length can not be too long, otherwise it will be a great amount of calculation. This is a major disadvantage of this method.
(3) of the possible solutions. To simplify the calculations, you can search only in the Alto area, even so, all solutions are still 10 times as much as the power of 21, but taking into account the heuristic information, we only need to search for a small part of the solution on it.
2. The stochastic input + SOFM clusters.
One of the drawbacks of the above search method is not creative, the lack of change, I think a very important point is the choice between the best and feasibility feasibility. So, is there a way to enhance the optimal solution of it?
In order to increase the length of the song, the cluster is a good method. Clustering is the similar objects into different groups or more subsets by means of static classification, so let similar in the same subset of members of the object has some properties.
Description:
(1) random notes can be generated randomly.
(2) "similar" may be defined near the frequency analysis may be obtained of a large number of music similarity table.
(3) and the above-mentioned search is not the same, since this input method is random, the output is also full of changes, avoiding monotonous, and a single low computational complexity, can simulate human emotions. Less than that, not necessarily the result of this clustering of human taste, it also requires a lot of computing, from which to choose our favorite solution.
3. Genetic Algorithms.
Evolutionary genetic algorithm learning biological nature, such processes are: the combination of notes arranged into a set of genes encoding a group selected from the parent population, the parent varied by generating a random stochastic perturbation, by picking the best fitness function individual, group and join the next evolution, until you find a feasible solution.
Description:
(1) Code: mapping the gene sequence of notes organism.
(2) random disturbance: After each generation evolution, random disturbance, random notes on the cross (reorganization) and mutation (change a bit in the notes), to diversify solution, so far as possible to achieve the overall best.
(3) options: an evolution after evolution by selecting the fitness function, to find the best adapted to the individual environment in question.
(4) the fitness function: evaluation of the merits of the solution, this is in the "sweet degree" song, reflected mathematically can make different notes, the chord is connected to, and may be similar masterpiece fragments and the like.
in conclusion
By reading these papers, algorithmic composition is not difficult, Sony, Google and other companies have
to create some music, and I think after listening to mimic existing music is easy to achieve, but whether people
when they start is how I think, music performance than syntactic rules can capture something much more. In a very
over a long period of time, a computer program composer to compose a musical piece not produce new beauty. Not just sound
music, all works of art, for us, is a kind of expression. It has to show content and emotion is much
more than just the rules and techniques that can be captured. It must come from an experience, but also in terms of this being
that the machine can not be achieved. But I believe that through the neural network model, machine learning and artificial intelligence algorithms
evolving to create music will be getting better and better.
Author: Get Smart Writing