About vctk datasets

1. download vctk Data sets （ Download path ：https://datashare.ed.ac.uk/download/DS_10283_3443.zip）
2. vctk Data set understanding ：
   1. CSTR VCTK The corpus includes 110 Voice data from English speakers using different accents . Each speaker reads about 400 A sentence , These sentences are from a newspaper 、rainbow The article and an inspirational paragraph for voice stress files .
   2. The text is selected according to the greedy algorithm , Greedy algorithm can increase context and voice coverage .
   3. All voice data are recorded using the same recording settings ： An omnidirectional microphone （DPA 4035） And a small diaphragm condenser microphone , The bandwidth is very wide （Sennheiser MKH 800）, Sampling frequency is 96kHz,24 position , Located in the semi anechoic room of the University of Edinburgh .
   4. All records are converted to 16 position , Downsampling to 48 kHz
   5. This corpus was originally used to base on HMM Text to speech synthesis system , Especially based on speaker adaptation HMM Voice synthesis of , The synthesis uses the average speech model of multiple speakers and speaker adaptation technology . The corpus is also applicable to DNN Multi spoken human language synthesis system and waveform modeling .** The idea here and PCA The idea of extracting face features and averaging faces to synthesize a given face is similar **
3. VCTK There are several variants of corpus ：
   1. Voice enhancement ： For training speech enhancement algorithms and TTS Model noise speech database , Audio is artificially directed to VCTK Various types of noise are added ：http://dx.doi.org/10.7488/ds/2117
   2. Reverberation voice database , Used to train speech de reverberation algorithm and TTS Model ,VCTK Various types of reverberation have been artificially added in http://dx.doi.org/10.7488/ds/1425
   3. For training speech enhancement algorithms and TTS Model noise reverberation speech database http://dx.doi.org/10.7488/ds/2139
   4. Equipment records VCTK, among VCTK The speech signal of the corpus is played back , And use relatively cheap consumer equipment to re record in the office environment http://dx.doi.org/10.7488/ds/2316
   5. Microsoft Scalable noisy speech data set （MS-SNSD）https://github.com/microsoft/MS-SNSD
   6. ASV And anti deception ：
      1. Deception and anti deception （SAS） corpus , It is a collection of synthetic speech signals produced by nine technologies , Two of them are speech synthesis , Seven are voice conversion . All of these are using VCTK Corpus construction .http://dx.doi.org/10.7488/ds/252
      2. Automated speaker verification deception and countermeasure challenges （ASVspoof 2015） database . The database is composed of synthetic speech signals generated by ten technologies , It has been used in the first automatic speaker verification deception and challenge confrontation （ASVspoof 2015）http://dx.doi.org/10.7488/ds/298
      3. ASVspoof 2019： The third automatic speaker verification deception and countermeasure challenge database . The database has been used for the third automatic speaker verification deception and countermeasure challenge （ASVspoof 2019）https://doi.org/10.7488/ds/2555
      4. To use the corpus, you need to add references ：

         Christophe Veaux,  Junichi Yamagishi, Kirsten MacDonald, "CSTR VCTK Corpus: English Multi-speaker Corpus for CSTR Voice Cloning Toolkit",   The Centre for Speech Technology Research (CSTR), University of Edinbur