我想做一个语音识别算法。我有一个wav文件，里面有+-20分钟的演讲。我把它读入numpy数组，每1024个值的块是一行。由于wave模块文件的readframes方法提供的所有块都具有相同的长度，因此使用numpy.padd函数填充一些行，以使数组具有均匀的形状。这些空格只会被添加到数组后面，因此不会导致以下情况:
我注意到数组中有一些列只包含零。这些列总是成对出现，并且总是由包含正常值的两列分隔。第六列似乎是这个模式的一个例外:它有时也包含1。这些栏目是真实录制的演讲，还是我的电脑出于某种原因把它们夹在中间?知道这一点很重要，因为我不想让我的算法在计算机生成的值上进行训练。我应该删除这些值还是保留它们更好?数组不再需要作为音频播放。
下面是数组中的一个示例。我不能附加完整的数组，因为它太大了。更完整的版本可以在这里找到。
[78, 1, 0, 0, 79, 1, 0, 0, 12, 1, 0, 0, 185, 0, 0, 0, 177, 0, 0, 0, 28, 1, 0, 0, 245, 1, 0, 0, 38, 3, 0, 0, 106, 4, 0, 0, 81, 5, 0, 0, 148, 5, 0, 0, 74, 5, 0, 0, 168, 4, 0, 0, 229, 3, 0, 0, 83, 3, 0, 0, 31, 3, 0, 0, 26, 3, 0, 0, 33, 3, 0, 0, 40, 3, 0, 0, 22, 3, 0, 0, 246, 2, 0, 0, 211, 2, 0, 0, 136, 2, 0, 0, 240, 1, 0, 0, 247, 0, 0, 0, 176, 255, 0, 0, 97, 254, 0, 0, 69, 253, 0, 0, 131, 252, 0, 0, 54, 252, 0, 0, 81, 252, 0, 0, 188, 252, 0, 0, 79, 253, 0, 0, 207, 253, 0, 0, 48, 254, 0, 0, 97, 254, 0, 0, 73, 254, 0, 0, 6, 254, 0, 0, 175, 253, 0, 0, 90, 253, 0, 0, 58, 253, 0, 0, 73, 253, 0, 0, 101, 253, 0, 0, 132, 253, 0, 0, 147, 253, 0, 0, 164, 253, 0, 0, 199, 253, 0, 0, 224, 253, 0, 0, 8, 254, 0, 0, 97, 254, 0, 0, 228, 254, 0, 0, 163, 255, 0, 0, 138, 0, 0, 0, 89, 1, 0, 0, 251, 1, 0, 0, 45, 2, 0, 0, 157, 1, 0, 0, 95, 0, 0, 0, 161, 254, 0, 0, 185, 252, 0, 0, 56, 251, 0, 0, 134, 250, 0, 0, 226, 250, 0, 0, 51, 252, 0, 0, 246, 253, 0, 0, 175, 255, 0, 0, 208, 0, 0, 0, 240, 0, 0, 0, 96, 0, 0, 0, 124, 255, 0, 0, 119, 254, 0, 0, 223, 253, 0, 0, 243, 253, 0, 0, 120, 254, 0, 0, 77, 255, 0, 0, 254, 255, 0, 0, 253, 255, 0, 0, 63, 255, 0, 0, 224, 253, 0, 0, 61, 252, 0, 0, 27, 251, 0, 0, 35, 251, 0, 0, 180, 252, 0, 0, 154, 255, 0, 0, 1, 3, 0, 0, 6, 6, 0, 0, 189, 7, 0, 0, 116, 7, 0, 0, 111, 5, 0, 0, 118, 2, 0, 0, 116, 255, 0, 0, 133, 253, 0, 0, 55, 253, 0, 0, 124, 254, 0, 0, 17, 1, 0, 0, 19, 4, 0, 0, 87, 6, 0, 0, 42, 7, 0, 0, 53, 6, 0, 0, 195, 3, 0, 0]
[240, 0, 0, 0, 235, 254, 0, 0, 155, 254, 0, 0, 87, 0, 0, 0, 80, 3, 0, 0, 25, 6, 0, 0, 137, 7, 0, 0, 253, 6, 0, 0, 146, 4, 0, 0, 57, 1, 0, 0, 35, 254, 0, 0, 73, 252, 0, 0, 56, 252, 0, 0, 185, 253, 0, 0, 251, 255, 0, 0, 60, 2, 0, 0, 204, 3, 0, 0, 18, 4, 0, 0, 15, 3, 0, 0, 21, 1, 0, 0, 137, 254, 0, 0, 95, 252, 0, 0, 99, 251, 0, 0, 137, 251, 0, 0, 210, 252, 0, 0, 46, 255, 0, 0, 161, 1, 0, 0, 40, 3, 0, 0, 102, 3, 0, 0, 95, 2, 0, 0, 127, 0, 0, 0, 92, 254, 0, 0, 160, 252, 0, 0, 18, 252, 0, 0, 216, 252, 0, 0, 133, 254, 0, 0, 181, 0, 0, 0, 136, 2, 0, 0, 248, 2, 0, 0, 213, 1, 0, 0, 130, 255, 0, 0, 202, 252, 0, 0, 199, 250, 0, 0, 241, 249, 0, 0, 56, 250, 0, 0, 121, 251, 0, 0, 15, 253, 0, 0, 72, 254, 0, 0, 235, 254, 0, 0, 188, 254, 0, 0, 180, 253, 0, 0, 61, 252, 0, 0, 191, 250, 0, 0, 167, 249, 0, 0, 108, 249, 0, 0, 33, 250, 0, 0, 144, 251, 0, 0, 106, 253, 0, 0, 20, 255, 0, 0, 227, 255, 0, 0, 169, 255, 0, 0, 173, 254, 0, 0, 82, 253, 0, 0, 3, 252, 0, 0, 69, 251, 0, 0, 115, 251, 0, 0, 146, 252, 0, 0, 109, 254, 0, 0, 137, 0, 0, 0, 64, 2, 0, 0, 28, 3, 0, 0, 241, 2, 0, 0, 233, 1, 0, 0, 148, 0, 0, 0, 143, 255, 0, 0, 60, 255, 0, 0, 183, 255, 0, 0, 185, 0, 0, 0, 203, 1, 0, 0, 149, 2, 0, 0, 208, 2, 0, 0, 97, 2, 0, 0, 127, 1, 0, 0, 103, 0, 0, 0, 66, 255, 0, 0, 74, 254, 0, 0, 172, 253, 0, 0, 121, 253, 0, 0, 158, 253, 0, 0, 218, 253, 0, 0, 242, 253, 0, 0, 207, 253, 0, 0, 97, 253, 0, 0, 191, 252, 0, 0, 91, 252, 0, 0, 144, 252, 0, 0, 82, 253, 0, 0, 100, 254, 0, 0, 122, 255, 0, 0, 34, 0, 0, 0]
[1, 0, 0, 0, 65, 255, 0, 0, 89, 254, 0, 0, 151, 253, 0, 0, 47, 253, 0, 0, 69, 253, 0, 0, 221, 253, 0, 0, 237, 254, 0, 0, 76, 0, 0, 0, 166, 1, 0, 0, 187, 2, 0, 0, 133, 3, 0, 0, 3, 4, 0, 0, 86, 4, 0, 0, 179, 4, 0, 0, 47, 5, 0, 0, 198, 5, 0, 0, 96, 6, 0, 0, 197, 6, 0, 0, 210, 6, 0, 0, 145, 6, 0, 0, 37, 6, 0, 0, 217, 5, 0, 0, 226, 5, 0, 0, 73, 6, 0, 0, 35, 7, 0, 0, 86, 8, 0, 0, 105, 9, 0, 0, 11, 10, 0, 0, 252, 9, 0, 0, 222, 8, 0, 0, 224, 6, 0, 0, 151, 4, 0, 0, 73, 2, 0, 0, 111, 0, 0, 0, 152, 255, 0, 0, 125, 255, 0, 0, 130, 255, 0, 0, 105, 255, 0, 0, 214, 254, 0, 0, 135, 253, 0, 0, 223, 251, 0, 0, 108, 250, 0, 0, 144, 249, 0, 0, 117, 249, 0, 0, 255, 249, 0, 0, 244, 250, 0, 0, 239, 251, 0, 0, 101, 252, 0, 0, 38, 252, 0, 0, 99, 251, 0, 0, 105, 250, 0, 0, 195, 249, 0, 0, 239, 249, 0, 0, 4, 251, 0, 0, 216, 252, 0, 0, 218, 254, 0, 0, 61, 0, 0, 0, 161, 0, 0, 0, 19, 0, 0, 0, 227, 254, 0, 0, 190, 253, 0, 0, 69, 253, 0, 0, 179, 253, 0, 0, 209, 254, 0, 0, 21, 0, 0, 0, 250, 0, 0, 0, 35, 1, 0, 0, 76, 0, 0, 0, 162, 254, 0, 0, 204, 252, 0, 0, 104, 251, 0, 0, 239, 250, 0, 0, 177, 251, 0, 0, 127, 253, 0, 0, 162, 255, 0, 0, 42, 1, 0, 0, 95, 1, 0, 0, 246, 255, 0, 0, 23, 253, 0, 0, 154, 249, 0, 0, 220, 246, 0, 0, 235, 245, 0, 0, 32, 247, 0, 0, 33, 250, 0, 0, 214, 253, 0, 0, 203, 0, 0, 0, 227, 1, 0, 0, 234, 0, 0, 0, 137, 254, 0, 0, 202, 251, 0, 0, 251, 249, 0, 0, 42, 250, 0, 0, 77, 252, 0, 0, 154, 255, 0, 0, 24, 3, 0, 0, 162, 5, 0, 0, 107, 6, 0, 0, 139, 5, 0, 0, 154, 3, 0, 0]
文件似乎包含两个不同的声道。第二个声音通道没有录制，因为我只有一个麦克风。这导致该通道默认为全零。因为.wav文件使用两个比特来记录每个通道，这导致两个地方被零填充。然后wave模块将所有不同的通道组合成一个块，每个块包含2个正确的位和2个不正确的(0)位，并将这些帧中的100个(我的块大小)组合成一个数组，从而传递给定的数组。
我可以在没有任何音频质量损失的情况下完美地从数组中删除零。但是，我还必须将通道数量减少到1，因为我刚刚删除了第二个通道。