Voiceprint Recognition·General Chapter

Voiceprint recognition of interest " who said ," to solve biological identification and recognition ; and voice recognition "say what" Caring for solving recognition of what is said.

1. Theory

1.1 Basics of Voiceprint Recognition

• Simple knowledge of waves (sound waves as an example)

• Audio characteristics (1): Time domain diagram

• Audio features (2): time domain diagram, spectrogram, and spectrogram (time-spectrogram)

• Audio features (3): Summary of various waveform images

• The principle behind "sound" (1): sound waves, sound collection, sound storage

• The principle behind "sound" (2): sampling, quantization and coding

• The principle behind "sound" (3): A summary of the basic concepts and processes of audio signal processing

• Overview of Voiceprint Recognition (1) First acquaintance

• Overview of Voiceprint Recognition (2) Principle and Process of Voiceprint Recognition

• Overview of Voiceprint Recognition (3) Voiceprint Recognition System

• Summary of Voiceprint Recognition Technology (1): Voiceprint Modeling Technology

1.2 Voiceprint recognition algorithm

• Voiceprint recognition: the principle of x-vector feature extraction
• Kaldi speaker recognition: plda adaptive based on x-vector

1.3 Introduction to Voiceprint Recognition Data

Introduction to common data sets for voiceprint recognition

2. Resources

2.1 Data (Chinese/English)

(1) Chinese data set
SLR33
SLR85
SLR82
AISHELL-2
SLR18

(2) The foreign data set
Voxceleb2
can be downloaded by clicking here (Note: Decompression method)
Supporting paper: Paper sharing VoxCeleb2: Deep Speaker Recognition

2.2 Tools

Left

• Kaldi Voiceprint Recognition·Resource Summary

• Kaldi Voiceprint Recognition·General Chapter

  Python + kaldi
  PYTORCH-KALDI speech recognition toolkit

TensorFlow

TensorFlow-based Deep Speaker

PyTorch

PyTorch-based Deep Speaker

Hard

Speaker Recognition

2.3 Resource summary

1. Explain the principle, evolution and application selection of audio codec in detail -a very comprehensive series of audio related series!
2. Speech Recognition (8)-Voiceprint Recognition, Geography
3. Exploring Xiaobai Voiceprint Recognition (Speaker Recognition)
4. 20190510 Speech recognition resource finishing
5. iamxiaoyubei /Voice-Tech-Study
6. Kaldi/speech recognition ASR/voiceprint recognition SRE/resource summary -summarizes many comprehensive speech recognition/voiceprint recognition resources

Voiceprint recognition application

• The application practice of voiceprint recognition -Dr. Li's knowledge column, well written!
• Leon Jin’s voiceprint/ASR/diarization/Kaldi fan -you can see the answer from the big guy in Zhihu , maybe there will be new gains

Voiceprint recognition learning path

• Sorting out the information of the speaker recognition/voiceprint recognition learning path, learning voiceprint recognition from zero -it's awesome! If you have time and want to study systematically, you can refer to this learning path: GMM-UBM -> JFA -> Ivector-PLDA -> DNN embeddings -> E2E

2.4 Voiceprint recognition Daniel

• Wang Yun
• Wang Quan -This is very powerful, with introductions of many theories and tools, as well as the project maintained by Mr. Wang Quan, and the new book: "Voiceprint Technology: From Core Algorithms to Engineering Practice"

3. Actual combat: theory to code

data

Speech library analysis and
evaluation method of audio quality

index

• Basic indicators: FAR, FRR
  face recognition model evaluation indicators: a complete review
• How to obtain the best effect threshold for ERR
  voiceprint recognition (by calculating ERR)?

test

• Kaldi project test (1) Smoothly extract features and calculate similarity scores

Code

papers with code: Speaker Verification + papers + code! ! !

Voiceprint recognition project

1. Python + tensorflow: Overview of voiceprint recognition + a simple model implemented by tensorflow ( https://github.com/RDShi/voiceprint )

2. Python + Keras: Chinese and English voiceprint recognition based on Kersa

3. [Engineering] Application of in-depth speaker and production of voiceprint recognition system (with complete code and data set)

4. Use tensorflow and densenet neural network to realize spectrogram voiceprint recognition, that is, speaker recognition

   Supporting github address: https://github.com/jcfszxc/Project

reference:

1. Speech Recognition (8)-Voiceprint Recognition, Geography
2. Voiceprint recognition knowledge collation