ganga4364/stt_tibetan_dialects_data

Tibetan ASR Dataset

This dataset is designed for Automatic Speech Recognition (ASR) tasks with Tibetan speech data. It contains samples from various departments, and the data is split into training, validation, and test sets.

Dataset Overview

The dataset contains audio files in Tibetan, which have been grouped into different departments (STT_AM, STT_AM_AB, STT_KH, STT_KH_AB). It also includes transcripts in both Unicode and Wylie transliteration formats. The dataset is split into three subsets: train, validation, and test.

• Train: 112,870 samples
• Validation: 2,000 samples
• Test: 8,000 samples

Department-wise Statistics

Department	Number of Audio Files	Total Audio Length (hours)
STT_AM	26,736	28.19
STT_AM_AB	23,121	24.86
STT_KH	43,946	46.69
STT_KH_AB	30,272	34.07
Total	123,075	133.82

---

Dataset Features

• file_name: The name of the file.
• uni: Unicode transcription of the audio.
• wylie: Wylie transliteration of the audio.
• url: URL to download the audio file.
• dept: Department of the dataset (e.g., STT_AM, STT_AM_AB).
• grade: Grade of the audio sample.
• char_len: Length of the transcript (in characters).
• audio_len: Duration of the audio sample (in seconds).