TIE_shorts / README.md
raianand's picture
Update README.md
28c53e2 verified
metadata
license: apache-2.0
task_categories:
  - automatic-speech-recognition
  - text-to-speech
language:
  - en
pretty_name: Technical Indian English
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: validation
        path: data/validation-*
dataset_info:
  features:
    - name: audio
      struct:
        - name: array
          sequence:
            sequence: float32
        - name: path
          dtype: string
        - name: sampling_rate
          dtype: int64
    - name: split
      dtype: string
    - name: ID
      dtype: string
    - name: Transcript
      dtype: string
    - name: Normalised_Transcript
      dtype: string
    - name: Speech_Duration_seconds
      dtype: float64
    - name: Speaker_ID
      dtype: int64
    - name: Gender
      dtype: string
    - name: Caste
      dtype: string
    - name: Year_Class
      dtype: string
    - name: Speech_Class
      dtype: string
    - name: Discipline_Group
      dtype: string
    - name: Native_Region
      dtype: string
    - name: Topic
      dtype: string
  splits:
    - name: train
      num_bytes: 12626734601
      num_examples: 7884
    - name: test
      num_bytes: 1548446759
      num_examples: 986
    - name: validation
      num_bytes: 1576842184
      num_examples: 986
  download_size: 15746227296
  dataset_size: 15752023544

Dataset Card for TIE_Shorts

Table of Contents

Dataset Description

Dataset Summary

TIE_shorts is a derived version of the Technical Indian English (TIE) dataset, a large-scale speech dataset (~ 8K hours) originally consisting of approximately 750 GB of content sourced from the NPTEL platform. The original TIE dataset contains around 9.8K technical lectures in English delivered by instructors from various regions across India, with each lecture averaging about 50 minutes. These lectures cover a wide range of technical subjects and capture diverse linguistic features characteristic of Indian English.

The TIE_shorts version (~ 70 hours audio and 600K ground-truth tokens) was created to facilitate efficient training and usage in speech processing tasks by providing shorter audio samples. In TIE_shorts, consecutive audio snippets from the original dataset were merged based on timestamps, with a condition that the final merged audio should not exceed 30 seconds in duration. This process results in 25–30 second audio clips, each accompanied by a corresponding ground-truth transcript. This approach retains the linguistic diversity of the original dataset while significantly reducing the size and complexity, making TIE_shorts ideal for Automatic Speech Recognition (ASR) and other speech-to-text applications. As the dataset consists of approximately 9.8K files spoken by 331 speakers from diverse demographics across the Indian population, it is also well-suited for speaker identification and text-to-speech (TTS) training applications.

Example usage

The TIE_Shorts dataset provides labeled audio data with metadata, including fields like Speaker ID, Gender, Caste, Native Region, and more. You can load the dataset with different configurations to access specific data subsets.:

To load the entire TIE_Shorts dataset, use the following code:

from datasets import load_dataset

tie_shorts = load_dataset("raianand/TIE_shorts")

To load only a specific split (such as train, test, or validation), use:

tie_shorts_train = load_dataset("raianand/TIE_shorts", split="train")
tie_shorts_test = load_dataset("raianand/TIE_shorts", split="test")
tie_shorts_validation = load_dataset("raianand/TIE_shorts", split="validation")

Inference using Open AI Whisper model, :


from transformers import WhisperProcessor, WhisperForConditionalGeneration


# load model and processor
processor = WhisperProcessor.from_pretrained("openai/whisper-base")
model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")

sample = tie_shorts_test[0]["audio"]
input_features = processor(sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features 

# generate token ids
predicted_ids = model.generate(input_features)
# decode token ids to text
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
print(transcription)
['the first time and therefore, because I find a lot of them have plagiarized therefore, I will not deduct or make any punishment for plagiarism then what the teacher tends to be arriving it as is arriving']

Dataset Structure

Data Instances

{
ID: GGlaqd17Ctg,
audio: {'array': [[-0.05644894391298294, -0.07796351611614227 ]],'sampling_rate':16000},
split: train ,
Transcript: So, and various details are listed there in the map it will not be very clear right now in this video screen. But I will advise you to purchase the map or go to a laboratory or somewhere where you can have a map.,
Normalised_Transcript: so and various details are listed there in the map it will not be very clear right now in this video screen but i will advise you to purchase the map or go to a laboratory or somewhere where you can have a map,
Gender: M,
Speaker_ID: 74,
Native_Region: NORTH,
Caste: UR,
Speech_Duration_seconds: 16.88,
Year_Class: LES_2000,
Speech_Class: FAST,
Discipline_Group: Engineering,
Topic: Lecture 1 Surveying,
}

Data Fields

Data Fields for TIE_Shorts

The dataset has the following structure:

  • audio_id (string) - The unique identifier for each audio segment.
  • audio (dict) - A dictionary containing the following fields related to the audio:
    • array (numpy.ndarray) - A NumPy array representing the decoded audio waveform. For brevity, only the first few samples are shown.
    • sampling_rate (int) - The sampling rate of the audio, typically 16000 Hz for this dataset.
  • raw_text (string) - The original, unmodified (orthographic) transcription of the audio segment.
  • normalized_text (string) - The normalized transcription of the audio segment, which is typically cleaned and adjusted for clarity.
  • gender (string) - The gender of the speaker (e.g., "M", "F").
  • speaker_id (string) - A unique identifier for the speaker.
  • caste (string) - The caste group of the speaker, (RES: Reserved Category, UR: Unreserved Category)
  • speech_duration_seconds (float) - The duration of the speech in seconds.
  • year_class (string) - The academic year and class the speaker belongs to (e.g., LES_1980: Lecturers with PhD before 1980, LES_1990: Lecturers with PhD between 1980 to 1990, LES_2000: Lecturers with PhD between 1990 to 2000, GRT_2000:Lecturers with PhD post 2000 ).
  • speech_class (string) - The classification of speech rate, e.g., "SLOW", "AVG", "FAST".
  • native_region (string) - Indian region to which speaker belongs to. ("WEST","EAST","NORTH","SOUTH")
  • discipline_group (string) - The speaker's discipline or academic field (e.g., "Engineering", "Non-Engineering").
  • topic (string) - The topic of the lecture or speech given by the speaker.

Source Data

The audio data and corresponding ground-truth transcripts are sourced from NPTEL Platform

Licensing Information

The dataset is distributed under Attribution-ShareAlike 2.0 Generic (CC BY-SA 2.0).

Citation Information

Please cite this paper:

@inproceedings{rai2024deep,
  title={A Deep Dive into the Disparity of Word Error Rates across Thousands of NPTEL MOOC Videos},
  author={Rai, Anand Kumar and Jaiswal, Siddharth D and Mukherjee, Animesh},
  booktitle={Proceedings of the International AAAI Conference on Web and Social Media},
  volume={18},
  pages={1302--1314},
  year={2024}
}