---

---

Download Edit

---

---

Step 1.- The audio is transcribed to notes by using a RNN model. You can find more information about this model here.

---

Step 2.- The notes are normalized, cleaned and adjusted to 4/4

---

Step 3.- Now we use a DNN model to predict the most likely genre from the melody. You can find more information about this model here.

The predicted genre is:

GENRE

The scores are:

ROCK: 0.0

  

DANCE: 0.0

  

JAZZ: 0.0

---

Step 4.- The melody is adapted to an instrument that is typically used in the predicted genre.

---

Step 5.- A temporal drum is generated.

---

Step 6.- A temporal bass is generated by expanding and joining the transcribed melody.

---

Step 7.- We need to form a trio that is the combinaton of MELODY + BASS + DRUMS. This is needed to predict a sequence. This is the current trio.

---

Step 8.- We use a RNN model that generates a trio that is the progression of the current trio. You can find more information about this model here.

---

Step 9.- We use the original melody, the generated bass and the random drum to generate a base trio.

---

Step 10.- Now we define the harmony, the first step is to calculate the musical scale from the melody.

The tonal scale that fits better is:

A

---

Step 11.- From that scale we choose a chord progression that fits better to the melody.

The chord progression that fits better is:

A, A, A, A

---

Step 12.- Now we use a RNN model that is able to generate an accompaniment for our main trio, adapted to the chord progression. You can find more information about this model here.

---

Step 13.- We also generate random accompaniment that fits to the chosen chord progression.

---

Step 14.- Finally, we create an interpolation of 6 steps between the chords adapted trio and random sequence. We use a RNN model that can be found here.

---

We appreciate your feedback, you can find the contact information here.