声音合成：使用 AS3 在频率之间插值
Posted
技术标签:
【中文标题】声音合成：使用 AS3 在频率之间插值【英文标题】：Sound synthesis: interpolate betweeen frequencies using AS3 【发布时间】：2016-10-26 00:03:49 【问题描述】：我有点迷茫，希望有人能对此有所了解。 出于好奇，我正在开发一个简单的软合成器/音序器。一些想法 取自家用电脑黄金时代流行的 .mod 格式。 目前它只是一个模型。从数组中读出注释 最多 64 个值，其中数组中的每个位置对应于十六分之一 笔记。到目前为止一切顺利，一切正常，旋律响起 正好。如果从一个音符过渡到另一个音符，就会出现问题。 例如f4 -> g#4。由于这是一个突然的变化，因此会出现明显的弹出/点击 声音。为了补偿我试图在不同频率之间进行插值 并开始编写一个简单的示例来说明我的想法并验证它 工作。
import flash.display.Sprite;
import flash.events.Event;
import flash.display.Bitmap;
import flash.display.BitmapData;
public class Main extends Sprite
private var sampleRate:int = 44100;
private var oldFreq:Number = 349.1941058508811;
private var newFreq:Number = 349.1941058508811;
private var volume:Number = 15;
private var position:int = 0;
private var bmp:Bitmap = new Bitmap();
private var bmpData:BitmapData = new BitmapData(400, 100, false, 0x000000);
private var col:uint = 0xff0000;
public function Main():void
if (stage)
init();
else
addEventListener(Event.ADDED_TO_STAGE, init);
private function init(e:Event = null):void
removeEventListener(Event.ADDED_TO_STAGE, init);
bmp.bitmapData = bmpData;
addChild(bmp);
for (var a:int = 0; a < 280; a++)
if (a == 140)
col = 0x00ff00;
newFreq = 415.26411519488113;
if (a == 180)
col = 0x0000ff;
oldFreq = oldFreq * 0.9 + newFreq * 0.1;
bmpData.setPixel(position, Math.sin((position) * Math.PI * 2 / sampleRate * oldFreq * 2) * volume + bmpData.height/2, col);
position++;
这将生成以下输出：
蓝点代表 349.1941058508811 赫兹的正弦波，红色代表 415.26411519488113 赫兹，绿点代表插值。 对我来说，看起来这应该可行！ 但是，如果我将这种技术应用于我的项目，结果就不一样了！ 事实上，如果我将输出渲染到一个波形文件，这些之间的转换 两个频率看起来像这样：
显然它使弹出更糟。什么可能是错的？ 这是我的（缩短的）代码：
import flash.display.*;
import flash.events.Event;
import flash.events.*;
import flash.utils.ByteArray;
import flash.media.*;
import flash.utils.getTimer;
public class Main extends Sprite
private var sampleRate:int = 44100;
private var bufferSize:int = 8192;
private var bpm:int = 125;
private var numberOfRows:int = 64;
private var currentRow:int = 0;
private var quarterNoteLength:Number;
private var sixteenthNoteLength:Number;
private var numOctaves:int = 8;
private var patterns:Array = new Array();
private var currentPattern:int;
private var songOrder:Array = new Array();
private var notes:Array = new Array("c-", "c#", "d-", "d#", "e-", "f-", "f#", "g-", "g#", "a-", "a#", "b-");
private var frequencies:Array = new Array();
private var samplePosition:Number = 0;
private var position:int = 0;
private var channel1:Object = new Object();
public function Main():void
if (stage)
init();
else
addEventListener(Event.ADDED_TO_STAGE, init);
private function init(e:Event = null):void
removeEventListener(Event.ADDED_TO_STAGE, init);
quarterNoteLength = sampleRate * 60 / bpm;
sixteenthNoteLength = quarterNoteLength / 2 / 2;
for (var a:int = 0; a < numOctaves; a++)
for (var b:int = 0; b < notes.length; b++)
frequencies.push(new Array(notes[b % notes.length] + a, 16.35 * Math.pow(2, frequencies.length / 12)));
patterns.push(new Array("f-4", "", "", "", "g#4", "", "", "f-4", "", "f-4", "a#4", "", "f-4", "", "d#4", "", "f-4", "", "", "", "c-5", "", "", "f-4", "", "f-4", "c#5", "", "c-5", "", "g#4", "", "f-4", "", "c-5", "", "f-5", "", "f-4", "d#4", "", "d#4", "c-4", "", "g-4", "", "f-4", "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", ""));
songOrder = new Array(0, 0);
currentRow = 0;
currentPattern = 0;
channel1.volume = .05;
channel1.waveform = "sine";
channel1.frequency = [0];
channel1.oldFrequency = [0,0,0,0];
channel1.noteTriggered = false;
updateRow();
var sound:Sound = new Sound();
sound.addEventListener(SampleDataEvent.SAMPLE_DATA, onSampleData);
sound.play();
private function updateRow():void
var tempNote:String = patterns[songOrder[currentPattern]][currentRow];
if (tempNote != "")
channel1.frequency = new Array();
if (tempNote.indexOf("|") == -1)
channel1.frequency.push(findFrequency(tempNote));
channel1.noteTriggered = true;
private function onSampleData(event:SampleDataEvent):void
var sampleData:Number;
for (var i:int = 0; i < bufferSize; i++)
if (++samplePosition == sixteenthNoteLength)
if (++currentRow == numberOfRows)
currentRow = 0;
if (++currentPattern == songOrder.length)
currentPattern = 0;
updateRow();
samplePosition = 0;
for (var a:int = 0; a < (channel1.frequency).length; a++ )
channel1.oldFrequency[a] = channel1.oldFrequency[a]*0.9+channel1.frequency[a]*0.1;
if ((channel1.frequency).length == 1)
sampleData = generate(channel1.waveform, position, channel1.oldFrequency[0], channel1.volume);
else
sampleData = generate(channel1.waveform, position, channel1.oldFrequency[0], channel1.volume);
sampleData += generate(channel1.waveform, position, channel1.oldFrequency[1], channel1.volume);
event.data.writeFloat(sampleData);
event.data.writeFloat(sampleData);
position++;
private function generate(waveForm:String, pos:Number, frequency:Number, volume:Number):Number
var retVal:Number
switch (waveForm)
case "square":
retVal = Math.sin((pos) * 2 * Math.PI / sampleRate * frequency) > 0 ? volume : -volume;
break;
case "sine":
retVal = Math.sin((pos) * Math.PI * 2 / sampleRate * frequency * 2) * volume;
break;
case "sawtooth":
retVal = (2 * (pos % (sampleRate / frequency)) / (sampleRate / frequency) - 1) * volume;
break;
return retVal;
private function findFrequency(inpNote:String):Number
var retVal:Number;
for (var a:int = 0; a < frequencies.length; a++)
if (frequencies[a][0] == inpNote)
retVal = frequencies[a][1];
break;
return retVal;
谢谢！ =)
【问题讨论】：
顺便说一句，如果您将 140 替换为 160，将 180 替换为 210，您的原始插值也存在不同相位的问题。 【参考方案1】：您错过了当您切换频率时，generate 中的pos 值失去了不变性，也就是说，Math.sin((pos) * Math.PI * 2 / sampleRate * frequency * 2) 在以不同频率运行时会给出非常不同的值。相反，您应该使用“相位”变量，该变量将从 0 运行到 1，然后返回 0 并像锯齿图一样再次转发，并且将按（当前频率）*（1/采样率）的值转发。所以错误是您将两个generate() 结果添加到一个sampleData （因为干扰，您显然不能这样做）并且您使用一个position 作为时间值来计算相位而不是累积相位。检查这种方法，它应该会更好一点：
private function generate(waveForm:String, var phase:Number, frequency:Number, volume:Number):Number
// "pos" changed to "phase". This also means that "generate" should be called once per sample
var retVal:Number;
switch (waveForm)
case "square":
retVal = Math.sin(phase * 2 * Math.PI) > 0 ? volume : -volume;
break;
case "sine":
retVal = Math.sin(phase * 2 * Math.PI ) * volume;
break;
case "sawtooth":
retVal = (2*Math.abs(2*phase-1)-1)* volume;
break;
phase+=frequency/sampleRate;// calculate new phase
if (phase>1.0) phase-=1.0; // normalize phase to 0..1
return retVal;
private function onSampleData(event:SampleDataEvent):void
var sampleData:Number;
for(var i:int=0;i<bufferSize;i++)
if (++samplePosition == sixteenthNoteLength)
// leaving this part as is, seems working
if (++currentRow == numberOfRows)
currentRow = 0;
if (++currentPattern == songOrder.length)
currentPattern = 0;
updateRow();
samplePosition = 0;
sampleData=0;
for (i=0;i</*channels.length*/1;i++)
// TODO convert "channel1" to an array
// sampleData+=generate(channels[i].waveform, channels[i].phase, channels[i].frequency, channels[i].volume);
sampleData+=generate(channel1.waveform, channel1.phase, channel1.frequency[0], channel1.volume);
event.data.writeFloat(sampleData);
event.data.writeFloat(sampleData);
事实上，您的通道应该转到一个单独的类，它将所有参数（相位、频率、波形、音量）放在一起，然后，当您需要它们进行采样时，您只需调用 channels[i].generateNextSample() 和无需所有麻烦的参数即可获得浮点数。另外，一个频道，一个频率，所以跳过那些“oldFrequency”的东西。
作为后续，Channel 类的草图：
public class Channel
public const WAVE_SINE:int=0;
public const WAVE_SQUARE:int=1;
public const WAVE_SAWTOOTH:int=2;
private var phase:Number=0;
private var currentVolume:Number=0;
public var volume:Number; // 0 to 1, should build a setter to normalize
public var frequency:Number=0;
public var waveform:int; // should also not allow changing this mid-play probably
public function Channel(v:Number=0,wf:int=WAVE_SINE,f:Number=0)
this.volume=v;
this.frequency=f;
this.waveform=wf;
phase=0;
currentVolume=0;
public function generateNextSample():Number ... // use the generate() code above to fill
public function reset():void currentVolume=0; phase=0; // POW
// rest to taste, enabled, active, whatever
使用示例：
var ch:Vector.<Channel>=new Vector.<Channel>();
ch.push(new Channel());
function onSampleData(e:SampleDataEvent):void
for (var j:int=0;j<8192;j++)
// here to input code that can alter channels' freqs, volumes etc
var sd:Number=0;
for (var i:int=ch.length-1; i>=0;i--) sd+=ch[i].generateNextSample();
e.data.writeFloat(sd);
e.data.writeFloat(sd);
【讨论】：
我认为这可能与阶段有关，但无法弄清楚如何解决它。无论如何，如果我将相位保持在不同频率之间，过渡看起来有点像这样：Goldwave 1 如果我在触发新音符时将其重置为 0，就像这样：Goldwave 2 我可以通过在新音符开始时在不同幅度之间进行插值，但也许有更好的方法。你有什么建议吗？感谢您提供有用的回答 Vesper！ 你可以将当前的音量插值到改变的音量，这样可以消除“Goldwave 1”中的凹凸，只要插值超过几十个样本即可。这需要volume 上的设置器和将currentVolume 缓慢调整到generateNextSample() 中指定的音量的代码。以上是关于声音合成：使用 AS3 在频率之间插值的主要内容，如果未能解决你的问题，请参考以下文章
如何在 iPhone/Mac 上使用 CoreAudio 合成声音