jainrishi2001 / nisp-dataset

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
Eg:

├── 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

This work is licensed under a Creative Commons Attribution 4.0 International License.