Audio Synth

Part 2: Render songs as audio

Your goal: create code to render your song model to an AudioBuffer so you can hear it.

This is where the abstract description of music in the Song class becomes actual audio.

/**
 * Renders all the notes in the piece to a normalized audio buffer.
 */
public AudioBuffer renderAudio() {
    ...
}

Sketch of the method:

  1. Create a new AudioBuffer.

    Compute the size of the buffer using the song’s duration. Pay attention to units!

    Hint about units

    There is a method in Song that gives you the duration in seconds. But AudioBuffer needs you to specify a duration in samples.

    There is already a method that converts seconds to samples.

    More of a hint

    The method that converts seconds to samples is in the Utils class.

    Just show me how to use it!

    Utils.convertSecondsToSamples(getDuration())

  2. For each note in the song:

    • Create a signal for the note by converting its pitch to a wavelength. (Use the class diagram you drew to help you.)

      Hint about unit conversion

      There is another method in Utils that will help you.

      Hint about creating the signal

      Look at your class diagram. How can you get from Note to Signal?

      More of a hint about creating the signal

      A Note has a Waveform. A Waveform can create a Signal if you specify the wavelength.

    • Tell your audio buffer to mix the signal into the buffer at the correct time and for the correct duration.

  3. Call normalize() on your audio buffer. This prevents clipping in the mixed audio.

    Aside: What is “clipping?”

    Digital audio playback always has a limited range of allowed amplitudes. In AudioBuffer, this range is -1…1. Audio outside this range gets “clipped:” the top of the wave gets flattened out so that the wave stays within the allowed bounds. The resulting audio is loud and distorted.

    It is OK if intermediate calculations go outside the allowed range, but to avoid clipping, software must ensure that the final result stays within range. That is what calling normalize() will do for you.

  4. Return the audio buffer.

Add this test to SongTest:

@Test
void renderAudio() {
    double sampleLen = 1.0 / AudioBuffer.SAMPLE_RATE;
    song.addNote(new Note(waveform0, 81, 2 * sampleLen, 4 * sampleLen));
    song.addNote(new Note(waveform1, 33, 4 * sampleLen, 5 * sampleLen));

    AudioBuffer audio = song.renderAudio();
    assertArrayEquals(
        new float[] { 0, 0, 0.25f, 0.25f, -0.5f, 1, -0.75f, 0.75f, -0.75f },
        audio.getSamples(),
        0.0001f);
}

Does that work? Try uncommenting the commented-out code in IntegrationTest, which checks all of the pieces working together.

Did those tests pass? Run the main method in AudioSynth! You should hear a nice E♭ flat major triad.

If that works, then comment out the simple 3-note song, and uncomment the code below that loads a song from a data file. You can use both kondo.csv and bach.csv for testing.

Enjoy the tunes! Then, when you’re ready, proceed to the exciting next step.

Next: Visualization