EMO-DB 数据集的 Speech 特征提取

数据集介绍

- EMO-DB数据集是由柏林工业大学录制的德语情感语音库
- 535条语音文件（一共有十名演员其中五男五女）其中语句内容包含日常生活用语的5个短句和5个长句，具有较高情感自由度，不包含某一特定情感倾向。采用16kHZ采样，16bit量化，并以WAV格式保存文件。
- 其中情绪组成：anger, neutral, fear, boredom, happiness, sadness, disgust.

主要是用到这个语音标签 audio 文件的XXXWa.wav中W就是情绪

提取相关特征

import feature
import os
import pickle
import tqdm
def extract_emodb(file_name:str):
    label = file_name.split('.')[0][-2]
    feature_27 = feature.extract_mfcc_feature(file_name)
    return (label,feature_27)


database_dir = "H:\dataset\EM-DB\wav"
feature_tuple =()
dir_file = os.listdir(database_dir)
for i in tqdm.tqdm(dir_file):
    file_path = os.path.join(database_dir,i)
    feature_tuple = feature_tuple+ (extract_emodb(file_path),)
with open('emo_db.pkl', 'wb+') as file:
    pickle.dump(feature_tuple,file)

工具

def read_file(filename):
    """
    input a filename and get wave data and time,framerate
    :param filename:
    :return:
    """
    file = wave.open(filename, 'r')
    params = file.getparams()
    nchannels, sampwidth, framerate, wav_length = params[:4]
    str_data = file.readframes(wav_length)
    wavedata = np.fromstring(str_data, dtype=np.short)

    time = np.arange(0, wav_length) * (1.0 / framerate)
    file.close()
    return wavedata, time, framerate
def extract_mfcc_feature(filename):
    """
    input a filename extract mfcc from audio
    :param filename:
    :return:
    """
    data, time, rate = read_file(filename)
    mel_spec = ps.logfbank(data, rate, nfilt=13)
    np.set_printoptions()
    time = mel_spec.shape[0]
    if time <= 300:
        part = mel_spec
        part = np.pad(part, ((0, 300 - part.shape[0]), (0, 0)), 'constant',
                      constant_values=0)
        mfcc = dct(part)
    else:
        begin = 0
        end = 300
        part = mel_spec[begin:end, :]
        mfcc = dct(part)
    return mfcc

总结

这个数据集还是比较简单，明了，听不懂。