处理帧时如何解决webrtcvad.Error: Error？

Question

import glob
import scipy.io.wavfile as wav
import pandas as pd
import numpy as np
import scipy
import librosa
import webrtcvad

def get_vector(sig,rate):
    vec=np.empty((1,3))
    start=0
    end=320

    while(sig.shape[0]>=end+160):
        vad = webrtcvad.Vad()
        vad.set_mode(2)

        res=vad.is_speech(sig[start:end].tobytes(),rate) #speech_probability
        zero_crosses = np.nonzero(np.diff(sig[start:end]>0))[0].shape[0]/0.02 # zero crosses
        f=scipy.fft(sig[start:end])
        f0=min(np.absolute(f))                                         # f0 frequency

        start=start+160
        end=end+160

        vec=np.vstack((vec,np.array([res,zero_crosses,f0],ndmin=2)))

    mfcc_feat=librosa.feature.mfcc(sig,rate,numcep=12,winlen=0.020)[0:vec.shape[0],:]  # mfcc 
    fbank=librosa.feature.melspectrogram(sig,rate,nfilt=5)[0:vec.shape[0],:] # log filterbank energies 
    mfcc_grad=np.gradient(mfcc_feat,axis=0)                            # mfcc first derivative

   final_feature=np.hstack((mfcc_feat,mfcc_grad,fbank,vec))

   return final_feature


df=pd.DataFrame()
for i in range(1,6):
    for file in glob.glob("Actor_0102/*.wav".format(i)):
        print(file)
        (sig,rate,) = librosa.load(file)

    # get mfcc
        mfcc_feat = librosa.feature.mfcc(sig,rate)

    # get filterbank energies
        fbank_feat = librosa.feature.melspectrogram(sig,rate)


        final_vector=get_vector(sig,rate)
        feed_dict={"Features":final_vector.astype(np.float64),"name":file.split('/')[-1].split(',')[0]}
        df=df.append(feed_dict,ignore_index=True)

df.to_csv("Mfccfeatures.csv")

我正在使用RAVDESS数据集并获得错误：

webrtcvad.Error: Error while processing frame

当我尝试创建多个音频文件的特性并将它们存储在CSV文件中时

Answer 1

我也面临着同样的错误信息，并最终解决了。

(我不知道你和我的问题是否相似.)

你已经检查过音频的采样率了吗？韦布尔特瓦德说

The WebRTC VAD only accepts 16-bit mono PCM audio, sampled at 8000, 16000, 32000 or 48000 Hz。

当我尝试48000 (最初是44100)的速率时，它起了作用。