Formed in 2009, the Archive Team (not to be confused with the archive.org Archive-It Team) is a rogue archivist collective dedicated to saving copies of rapidly dying or deleted websites for the sake of history and digital heritage. The group is 100% composed of volunteers and interested parties, and has expanded into a large amount of related projects for saving online and digital history.
Dual Voice Mixer for TIMIT Dataset
This project is intended to be used in conjunction with the Pytorch implementation of LAS and test the LAS model's recognition ability when background noise is introduced. The TIMIT dataset is modified to generate voices with background noise that can be tested with a trained LAS model.
Setup
TIMIT
-
Move the files in mix_timit to the LAS Pytorch directory.
-
TIMIT dataset folder must be in the same directory as timit_preprocess.sh and mixer.py
-
Run timit_preprocess.sh (should convert NIST .WAV to RIFF .wav)
-
Run mixer.py
- TIR and gender mixing can be adjusted by editing their respective lists
-
Run timit_preproccess_mixed.py
- Adjust TIR and gender list accordingly
-
Run test_timit_mixed.py to generate phoneme error rate results
Requirements
-
pysox: Mixes audio files
-
SoX: Converts NIST to RIFF and a requirement for pysox
-
NumPy: Calculates target-to-interference ratio
-
pandas: Saves testing data in .csv format