Dataset · openbhasha
Telugu transcribed speech with disfluencies and tags
openbhasha/rsml-bench
Nearly 16 hours of Telugu audio with detailed transcription, timecodes, and speech feature markup.
What's inside
The dataset includes 40 full audio recordings (16 kHz, mono WAV) from YouTube lectures and interviews under a CC BY license, with a total duration of 15.82 hours. Each file is accompanied by time-aligned RSML transcription indicating speakers and manually verified segmentation. Special tags are placed in the text for code-mixing, normalization, disfluency, and background voices.
What you can do with it
- Training and testing ASR models for the Telugu language in real-world conditions
- Recognizing spontaneous speech with disfluencies and language mixing
- Training speaker diarization and long audio segmentation systems
- Tasks
- Automatic speech recognition
- Languages
- Telugu
- Size
- up to 1K rows
- Downloads this month
- 31.7K
- Likes
- 0
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("openbhasha/rsml-bench")