Java 类名：com.alibaba.alink.operator.batch.audio.ExtractMfccFeatureBatchOp
Python 类名：ExtractMfccFeatureBatchOp

功能介绍

  • 从数据中提取 MFCC 特征。
  • 支持Alink Vector、一维或两维Alink FloatTensor格式的数据

    使用方式

    用于声学特征提取，通常与ReadAudioToTensor组件一起使用，连接在其后

    文献索引

    [1] Davis S, Mermelstein P. Comparison of parametric representations for monosyllabic word recognition in continuously spoken sentences[J]. IEEE transactions on acoustics, speech, and signal processing, 1980, 28(4): 357-366.

    参数说明

    | 名称 | 中文名称 | 描述 | 类型 | 是否必须？ | 取值范围 | 默认值 | | —- | —- | —- | —- | —- | —- | —- |

| sampleRate | 采样率 | 采样率 | Integer | ✓ | | |

| selectedCol | 选中的列名 | 计算列对应的列名 | String | ✓ | | |

| hopTime | 相邻窗口时间间隔 | 相邻窗口时间间隔 | Double | | | 0.032 |

| numMfcc | mfcc参数 | mfcc参数 | Integer | | | 128 |

| outputCol | 输出结果列 | 输出结果列列名，可选，默认null | String | | | null |

| reservedCols | 算法保留列名 | 算法保留列 | String[] | | | null |

| windowTime | 一个窗口的时间 | 一个窗口的时间 | Double | | | 0.128 |

| numThreads | 组件多线程线程个数 | 组件多线程线程个数 | Integer | | | 1 |

代码示例

Python 代码

以下代码仅用于示意，可能需要修改部分代码或者配置环境后才能正常运行！

  1. dataDir = "https://alink-test-data.oss-cn-hangzhou.aliyuncs.com/audio";
  3. df = pd.DataFrame([
  4.     ["246.wav"],
  5.     ["247.wav"]
  6. ])
  8. allFiles = BatchOperator.fromDataframe(df, schemaStr='wav_file_path string')
  9. SAMPLE_RATE = 16000
  11. readOp = ReadAudioToTensorBatchOp().setRootFilePath(dataDir) \
  12. .setSampleRate(SAMPLE_RATE) \
  13. .setRelativeFilePathCol("wav_file_path") \
  14. .setOutputCol("tensor") \
  15. .linkFrom(allFiles)
  17. mfccOp = ExtractMfccFeatureBatchOp() \
  18. .setSampleRate(SAMPLE_RATE) \
  19. .setSelectedCol("tensor") \
  20. .linkFrom(readOp)
  22. mfccOp.print()

Java 代码

  1. import com.alibaba.alink.operator.batch.BatchOperator;
  2. import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
  3. import com.alibaba.alink.testutil.AlinkTestBase;
  4. import org.junit.Test;
  6. public class ExtractMfccFeatureBatchOpTest extends AlinkTestBase {
  7.     @Test
  8.     public void testExtractMfccFeatureBatchOp() throws Exception {
  9.         String dataDir = "https://alink-test-data.oss-cn-hangzhou.aliyuncs.com/audio";
  10.         String[] allFiles = {"246.wav", "247.wav"};
  11.         int sampleRate = 16000;
  12.         String tensorName = "tensor";
  13.         String mfccName = "mfcc";
  14.         String wavFile = "wav_file_path";
  15.         BatchOperator source = new MemSourceBatchOp(allFiles, wavFile)
  16.                 .link(new ReadAudioToTensorBatchOp()
  17.                         .setRootFilePath(dataDir)
  18.                         .setSampleRate(sampleRate)
  19.                         .setRelativeFilePathCol(wavFile)
  20.                         .setDuration(2)
  21.                         .setOutputCol(tensorName)
  22.                 )
  23.                 .link(new ExtractMfccFeatureBatchOp()
  24.                         .setSelectedCol(tensorName)
  25.                         .setSampleRate(sampleRate)
  26.                         .setWindowTime(0.128)
  27.                         .setHopTime(0.032)
  28.                         .setNumMfcc(26)
  29.                         .setOutputCol(mfccName))
  30.                 .select(new String[]{wavFile, mfccName})
  31.                 .print();
  32.     }
  33. }

运行结果

| wav_file_path | mfcc | | —- | —- |

| 246.wav | FLOAT#59,26,1#48.78127 -32.02646 12.432438 … |

| 247.wav | FLOAT#59,26,1#-50.62911 -13.844937 24.176699 … |