Datasets:
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: audio
dtype: audio
- name: speechid
dtype: string
- name: speaker_id
dtype: string
- name: review_score
dtype: int64
- name: transcript
dtype: string
- name: category
dtype: string
- name: speaker_gender
dtype: string
- name: speaker_age
dtype: string
splits:
- name: train
num_bytes: 579920220.506
num_examples: 1541
download_size: 422956016
dataset_size: 579920220.506
license: cc-by-4.0
task_categories:
- automatic-speech-recognition
language:
- ml
pretty_name: SMC Malayalam Speech Corpus
size_categories:
- 1K<n<10K
Dataset Card for [msc]
Table of Contents
- Table of Contents
- Dataset Description
- Dataset Structure
- Dataset Creation
- Considerations for Using the Data
- Additional Information
Dataset Description
- https://smc.org.in
- https://gitlab.com/smc/msc-reviewed-speech
- https://blog.smc.org.in/malayalam-speech-corpus/
- Point of Contact: Kavya Manohar
Dataset Summary
- 1541 speech samples
- 75 speech contributors
- 1:38:16 hours of speech
- 482 unique sentences
- 1400 unique words
- 553 unique syllables
- 48 unique phonemes
For more detailed analysis see the python notebook provided here
Supported Tasks and Leaderboards
Automatic Speech Recognition system development, gender and age identification of speakers
Languages
Malayalam
Dataset Structure
- file_name
- speechid
- speaker_id
- review_score
- transcript
- category (optional speech category)
- speaker_gender (optionally self declared)
- speaker_age (optionally self declared)
Data Instances
Data Fields
Data Splits
So specific Splits
Dataset Creation
The speech data is collected from volunteer users who read and record their speech through a web application using their personal devices. The recorded speech is reviewed (upvote and downvote gives a score of +1 and -1 respectively) by other users. The review score is also published.
Curation Rationale
The recorded speech is reviewed (upvote and downvote gives a score of +1 and -1 respectively) by other users. The review score is also published.
Curation Rationale
Those speech samples with at least three positive reviews are included in this dataset.
Source Data
Initial Data Collection and Normalization
The speech data is collected from volunteer contributors who read and record their speech through a web application. The users optionally provide name, age and gender. There is no further verification. Sentences to read out are curated by MSC Admin. The speech samples are reviewed by other users.
Personal and Sensitive Information
Every speaker is identified by a unique alphanumeric id and age and gender are published if the speaker has voluntarily published them.
Considerations for Using the Data
Social Impact of Dataset
Read speech corpus, recorded in natural environments by the users.
Dataset Curators
Kavya Manohar
Licensing Information
CC-BY-SA 4.0