Punjabi Speech Recognition Data from Sikhism Educational YouTube Videos

A collection of 999 YouTube videos totaling approximately 180 hours of audio content, focused on Sikhism-based educational topics in the Punjabi language. The dataset was created by alvynabranches and uploaded to Hugging Face in September 2024. It consists of weakly labeled transcriptions that lack punctuation and may have missing words at the start or end of utterances.

Use Cases

• Fine-tuning speech-to-text models for Punjabi based on the described audio and text content.
• Training language models on Punjabi educational transcripts, given the text data mentioned.
• Analyzing speech patterns in educational content, based on the 180 hours of audio described.
• Studying domain-specific vocabulary in Sikhism educational materials, as indicated by the dataset's content focus.

Strengths

• Contains approximately 180 hours of Punjabi audio content.
• Focuses on a specific domain (Sikhism educational content) from 999 distinct YouTube sources.

Limitations

• Transcriptions are weakly labeled and lack punctuation.
• The creator notes that transcriptions may have missing words at the start or end.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

YouTube videos, primarily Sikhism-based educational content.

Collection Method

Downloaded from YouTube.

Freshness

Last updated 2024-09-12 13:20:18; freshness should be verified.