hsyecheng/audio-fingerprinting

Audio fingerprinting and recognition in JAVA

audio-fingerprinting's Issues

findComplexPeaks函数里寻找局部最大值点的逻辑似乎写错了

在FindPeaks.java的findComplexPeaks方法里有这样一段逻辑：

for(int k = 0; k < len; k ++){
float pi = data_power[k];
boolean add = true;
for(int j = 0; j < neighborRange; j ++) {
float pl,pr;

            if(k - neighborRange >= 0) {
                pl = data_power[k - neighborRange];
            }else pl = pi - 1;

            if(k + neighborRange < len) {
                pr = data_power[k + neighborRange];
            }else pr = pi - 1;

            if (pi < pl && pi < pr) {
                add = false;
            }
        }
        if(add) add(pi, k);
    }

===========================================
内层的循环变量j似乎是搜索的半径，然而在循环语句里都是用的neighborRange定值，这里似乎写错了。

Accuracy is missing

I wrapped the code in an android app.

• song files are mono channel 16 bit signed little endian encoded via ffmpeg.
• then I used it to create index on 50 wav files ~ 30 second each and stored them in mysql db.
• I used android's audio recorder to generate PCM audio chunked in 6 second segments.

findings

• it recognises static audio [ fan noise, people talking ]
• while playing audio it provides random match of data