CoolFace
Datasetpublic

shivdi1999/nagri-sound-dataset

Sylheti Language Learning – Audio Interaction Dataset Overview This dataset contains letter-level reference pronunciation audio samples designed for a multilingual Augmented Reality (AR) language-learning system. The system adapts its interface language dynamically based on user preference, while primarily aiming to teach and evaluate Sylheti pronunciation.The dataset is structured to support real-time pronunciation feedback, deterministic AR triggers, and… See the full description on the dataset page: https://huggingface.co/datasets/shivdi1999/nagri-sound-dataset.

sourceHugging Faceccupdated 9mo agoView on Hugging Face
0likes6downloads
Dataset Card

Sylheti Language Learning – Audio Interaction Dataset

Overview

This dataset contains letter-level reference pronunciation audio samples designed for a multilingual Augmented Reality (AR) language-learning system.

The system adapts its interface language dynamically based on user preference, while primarily aiming to teach and evaluate Sylheti pronunciation. The dataset is structured to support real-time pronunciation feedback, deterministic AR triggers, and multilingual extensibility using Unity and Python.

The dataset is currently focused on Sylheti (syl) alphabet pronunciations and is structured to support Bengali (bn) and additional languages in future versions.

Language Design

To avoid ambiguity between spoken input, learning target, and interface language, the dataset separates language roles as follows:

  • —spoken_language The language spoken in the recorded audio sample (e.g., syl, bn).
  • —target_language The language the system is teaching or evaluating against (currently Sylheti for all reference samples).
  • —ui_language The language used by the AR interface and feedback system (handled dynamically at runtime in Unity and intentionally left empty in the dataset).

This separation ensures clarity for:

  • —reviewers
  • —collaborators
  • —Unity XR logic
  • —Python-based evaluation pipelines

Dataset Structure

Each row in the dataset represents one reference pronunciation sample for a single alphabet letter.

Audio files are stored externally and referenced using relative paths.

Folder Structure

alphabetsaudio/ ├── syl/ │ ├── vowels/ │ │ ├── sylV01.wav │ │ └── ... │ └── consonants/ │ ├── syl_C01.wav │ └── ... ├── bn/ (future expansion) │ ├── vowels/ │ └── consonants/

Column Definitions

Column NameDescription
sample_idUnique identifier for a specific audio sample
letter_idCanonical identifier for the letter (language + vowel/consonant + index)
audio_fileRelative path to the .wav audio file
spoken_languageLanguage spoken in the audio sample
target_languageLanguage the system is teaching or evaluating
ui_languageLanguage used by the AR interface (runtime-controlled)
labelGround-truth label for reference audio (correct)
expected_responseExpected AR feedback (e.g., green_tick)
environmentRecording context (e.g., indoor, outdoor)
duration_msLength of the audio sample in milliseconds
noise_levelApproximate background noise level
user_age_groupAge group of the speaker
mic_typeMicrophone or device used for recording
confidence_scoreAnnotator confidence score (0–1 scale)
annotator_idAnonymous identifier of the annotator
spectrogram_pathOptional path to a spectrogram image (unused in v0.x)
unity_action_idUnity AR action triggered by this sample
gesture_expectedExpected user gesture in the AR experience

Letter Identification Logic

The letter_id encodes language, phonetic category, and index:

  • —Sylheti vowels: syl_V01, syl_V02, …
  • —Sylheti consonants: syl_C01, syl_C02, …
  • —Bengali vowels: bn_V01, bn_V02, …
  • —Bengali consonants: bn_C01, bn_C02, …

This distinction is intentional and required to support:

  • —different pronunciation evaluation logic for vowels vs consonants
  • —deterministic Unity AR behavior
  • —future curriculum and analytics design

Sample Identification Logic

sample_id

Format: <letterid><sampleindex> sylV01_001

  • —Each letter begins with _001
  • —Additional recordings of the same letter may use _002, _003, etc.
  • —sample_id represents a recording instance, while letter_id represents the linguistic concept

Intended Use

This dataset is intended for:

  • —Letter-level pronunciation evaluation
  • —AR-based language learning experiences
  • —Unity XR interaction logic
  • —Python-based audio similarity and scoring pipelines
  • —Multilingual educational research

It is not intended for:

  • —sentence-level ASR
  • —conversational speech modeling
  • —large-vocabulary speech recognition

Limitations

  • —Single reference speaker per letter in current versions
  • —Controlled recording conditions
  • —No sentence-level or word-level data
  • —Dialectal variation within Sylheti is not exhaustively represented

These limitations are deliberate design choices for early-stage AR pronunciation systems.

Ethical Considerations

  • —No personally identifiable information is included
  • —Speaker metadata is stored only in coarse categories
  • —Audio data is collected with informed consent for research and educational use
  • —Dataset is designed to support inclusive, non-extractive language learning applications

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

license: cc-by-4.0