SpecDox: 172-Hour Urdu-to-English Speech Translation Dataset

172 hours of authentic audio captures the pure, real-world complexity of the Pakistani accent and code-mixed Urdu-English speech. SpecDox provides highly accurate, structured English transcriptions for complex audio inputs, architected for Urdu-to-English speech translation and ASR. The dataset was created by Shzaib and last updated on Hugging Face in June 2026.

Use Cases

• Train Urdu-to-English speech translation models based on the 172-hour audio corpus.
• Develop Automatic Speech Recognition (ASR) systems for Pakistani-accented English based on the described real-world audio.
• Benchmark model performance on code-mixed speech based on the dataset's focus on Urdu-English code-switching.
• Fine-tune models for accent-specific speech recognition based on the dataset's capture of the Pakistani accent.

Strengths

• 172 hours of authentic audio data provides substantial volume for model training.
• Focus on pure, real-world complexity of Pakistani accent and code-mixed speech, as described.
• Provides highly accurate, structured English transcriptions for complex audio inputs, according to the description.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.
• License information is unavailable, which restricts clarity on permissible use.

Provenance

Source

Shzaib on Hugging Face.

Collection Method

Likely collected from real-world sources, as described as 'authentic' and not synthetically augmented.

Freshness

Last updated 2026-06-14 12:29:10; freshness should be verified.

Geography

Likely Pakistan, given the focus on the Pakistani accent.