https://github.com/mravanelli/pyspeechrev
This python code performs an efficient speech reverberation starting from a dataset of close-talking speech signals and a collection of acoustic impulse responses.
https://github.com/mravanelli/pyspeechrev
convolution data-contamination distant-speech-recognition impulse-response speech-recognition speech-reverberation
Last synced: 10 months ago
JSON representation
This python code performs an efficient speech reverberation starting from a dataset of close-talking speech signals and a collection of acoustic impulse responses.
- Host: GitHub
- URL: https://github.com/mravanelli/pyspeechrev
- Owner: mravanelli
- Created: 2017-11-26T20:42:34.000Z (over 8 years ago)
- Default Branch: master
- Last Pushed: 2020-05-30T16:01:08.000Z (almost 6 years ago)
- Last Synced: 2025-04-05T16:24:02.693Z (about 1 year ago)
- Topics: convolution, data-contamination, distant-speech-recognition, impulse-response, speech-recognition, speech-reverberation
- Language: Python
- Size: 2.12 MB
- Stars: 95
- Watchers: 9
- Forks: 25
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# pySpeechRev
This python code performs an efficient speech reverberation starting from a dataset of close-talking speech signals and a collection of acoustic impulse responses.
The reverberated signal y[n] is computed in the following way:
```
y[n]=x[n] * h[n]
```
where x[n] is the clean signal and * is the convolutional operator.
The script takes in input the following arguments:
- in_folder: folder where the original close-talk dataset is stored.
- out_folder: folder where the reverberated dataset will be stored.
- list.txt : it is a text file where each row should contain: original_wav_file IR_file.
Before run it, make sure you have all the needed python packages. In particular:
- pysoundfile: ``pip install pysoundfile``
- numpy
- scipy
Example:
```
python pySpeechRev.py clean_examples/ rev_examples/ list.txt
```
Note that to have meaningful and realistic results, both the impulse responses and the clean speech signal must be sampled at the **same sampling rate** (e.g., 16 kHz - 16 kHz).
# Reverberated TIMIT
To create a reverberated version of TIMIT do the following steps:
- Make sure you have the TIMIT dataset. If not, it can be downloaded from the LDC website (https://catalog.ldc.upenn.edu/LDC93S1).
- Change lst_TIMIT.txt according to the paths of your TIMIT Dataset
- Run:
```
python pySpeechRev.py $path_TIMIT $path_TIMIT_rev lst_TIMIT.txt
```
The current version of TIMIT has been contaminated with some high-quality impulse responses of the DIRHA-English Dataset [3].
Tested on:
Python 2.7, Ubuntu
This code has been used in the following papers (please cite them if you use this code):
[1] M. Ravanelli, P. Svaizer, M. Omologo, "Realistic Multi-Microphone Data Simulation for Distant Speech Recognition", in Proceedings of Interspeech 2016. https://arxiv.org/abs/1711.09470
[2] M. Ravanelli, M. Omologo, "Contaminated speech training methods for robust DNN-HMM distant speech recognition", in Proceedings of INTERSPEECH 2015. https://arxiv.org/abs/1710.03538
[3] M. Ravanelli, M. Omologo, "The DIRHA-English corpus and related tasks for distant-speech recognition in domestic environments", in Proceedings of ASRU 2015. https://arxiv.org/abs/1710.02560