sulabhkatiyar/ne-asr-trp-aug

NE ASR Augmented Dataset -- Kokborok (trp)

v2 (2026-05-29): rebuilt 3x speed-only

The previous v1 (5x) revision applied +/-1 semitone pitch-shift augmentation to the train split. trp (Kokborok) is a tonal language with 2 phonemic tones (H/L); pitch-shift augmentation corrupts the tonal contrasts the model must learn. v1 is therefore DEPRECATED and should not be used for training.

v2 specs:

• Augmentation: speed-perturb only (factors per configs/augmentation_config.yaml v2 spec: 0.9, 1.0, 1.1).
• Aug factor: 3x (was 5x).
• Train rows: 7,533 (= 2,511 x 3).
• Val / test: unchanged from source (no augmentation on eval splits).
• Source: sulabhkatiyar/ne-asr-trp.
• Tonality fix per .claude/reports/module6/tonal_audit.md (root-cause: configs/augmentation_config.yaml previously flagged trp.tonal: false; corrected to true on 2026-05-29 commit 06f9b14).
• Rebuilt by: scripts/augment_data.py --config configs/augmentation_config.yaml --lang trp --clean (Phase B.5.2).

---

Augmented automatic speech recognition dataset for Kokborok (trp), a Tibeto-Burman language spoken in Tripura, India.

Source

Augmented from sulabhkatiyar/ne-asr-trp (original transcribed speech data from the ARTPARK-IISc Vaani project).

Language Information

Property	Value
Language	Kokborok
ISO 639-3	trp
Family	Tibeto-Burman
Region	Tripura, India
Tonal	Yes
Tier	C (3.81h original data)

Dataset Statistics

• Original training samples: 2,511
• Augmented training samples: 7,533 (3x augmentation)
• Train shards: 6
• Estimated original duration: ~3.8 hours
• Estimated augmented duration: ~11.4 hours

Split	Samples
train	7,533
validation	274
test	279

Transformations Applied

Each original training sample produces 3 samples (1 original + 2 speed + 0 pitch):

• Speed perturbation: 0.9x, 1.1x (2 variants per sample)
• Pitch shift: Disabled (tonal language -- pitch shift would alter lexical meaning)
• Noise augmentation: Not applied

SpecAugment Parameters (for training, NOT in this dataset)

These parameters are consumed by the training script and are not baked into the audio files:

• mask_time_prob: 0.07
• mask_time_length: 10
• mask_feature_prob: 0.05
• mask_feature_length: 10
• layerdrop: 0.05

Full augmentation config: configs/augmentation_config.yaml

Dataset Format

• Audio: 16kHz mono WAV (stored as Parquet with audio bytes)
• Text: Transcriptions
• Features: audio, text, language, augmentation
• Augmentation labels: original, speed_0.9, speed_1.1

How to Use

from datasets import load_dataset

# Load the full dataset
ds = load_dataset("sulabhkatiyar/ne-asr-trp-aug")

# Load only the training split
train = load_dataset("sulabhkatiyar/ne-asr-trp-aug", split="train")

# Filter to only original (non-augmented) samples
original_only = train.filter(lambda x: x["augmentation"] == "original")

# Filter to a specific augmentation type
speed_09 = train.filter(lambda x: x["augmentation"] == "speed_0.9")

Original Data

• Source dataset: sulabhkatiyar/ne-asr-trp
• Project: ARTPARK-IISc Vaani
• License: CC-BY-4.0

Citation

If you use this dataset, please cite the Vaani project and acknowledge the augmentation pipeline.