Awesome
NISP-Dataset
NISP-A Multilingual Multi-accent Dataset for Speaker Profiling
This dataset contains speech recordings along with speaker physical parameters (height, weight, .. )as well as regional information and linguistic information.
There are a total of 345 speakers(219 male and 126 female). The dataset contains sentences that are taken out from newspapers. Each speaker has contributed about 4-5 minutes of data that includes recordings in both English and their mother tongue. The transcript for the text is provided in UTF-8 format.
For each speaker following parameters were collected. This information is present in the file " speaker_details.csv "
| Parameters | Value/Unit |
|---|---|
| Gender | M | F |
| Mother Tongue | any of the 5 languages |
| Can read Mother Tongue | YES/NO |
| Medium of Instruction | any of the 5 languages |
| Language spoken with friends | any of the 5 languages |
| Age | in years |
| Height | in cm |
| Waist size | in cm |
| Shoulder size | in cm |
| Weight | in kg |
| Native / Place where brought up | District and State |
| Current Place of residence | District and State |
The recordings were performed in separate sessions for native language and English.
There are five folders in "NISP" folder, where each native language folder has speech files and corresponding transcripts files in "Scripts" folder. The details of the folder is as follows
<br> Eg:
<br>
├── Hindi_master
│ ├── Hindi
│ ├── └── RECS
│ ├── English_Hindi
│ ├── └── RECS
│ ├── scripts
│ ├── └── Hindi_txt
│ ├── └── English_hindi_txt
For each Speaker ID the saving format of the file is like as follows :
for .wav files
(Native language)(Speaker ID)(language recorded)(gender)(Utterance ID).wav Example: Hin_0001_Eng_f_0000.wav
for .txt files (transcripts)
(Native language)(Speaker ID)(language recorded)_(gender).txt Example: Hin_0001_Eng_f.txt
The set of langauges recorded are named as follows:
| Language | Lang_ID |
|---|---|
| English | Eng |
| Hindi | Hin |
| Kannada | Kan |
| Malayalam | Mal |
| Tamil | Tam |
| Telugu | Tel |
The details of the Speakers and Train and Test splits are given in
total_spkrinfo.list -- this file has the details like "Speaker_ID, Gender, Mother_Tongue, Height (cm), Shoulder_size (cm), Waist_size(cm), Weight(Kg), Age(y), Native_State, and Native_District"
test_spkrID -- list of test speakers
train_spkrID -- list of train speakers
Note: Other lingusitic details and regional details are provided in individual files in each native language Master folder.
If any wav file missing in any of the languages but corresponding text file is available means treat that sentence is not recorded.
How to extract the data
cat RECS.tar.gz.a* > Complete.tar.gz
gzip -dc Complete.tar.gz | tar -xvzf -
Paper
- This Dataset paper is available in ArXiv with this link : https://arxiv.org/abs/2007.06021
- From IEEE Explorer : https://ieeexplore.ieee.org/document/9414349
- Poster and video can be accessed from IEEE sigport : https://sigport.org/documents/nisp-multi-lingual-multi-accent-dataset-speaker-profiling
- YouTube link : https://youtu.be/l4E5JfnyC78
<a rel="license" href="http://creativecommons.org/licenses/by/4.0/"><img alt="Creative Commons Licence" style="border-width:0" src="https://i.creativecommons.org/l/by/4.0/88x31.png" /></a><br />This work is licensed under a <a rel="license" href="http://creativecommons.org/licenses/by/4.0/">Creative Commons Attribution 4.0 International License</a>.