Tajik ASR Corpus: Multisource Speech Data for Automatic Speech Recognition

Tajik ASR Corpus v0 is a deduplicated collection for automatic speech recognition assembled from multiple sources. The dataset, created by Peacockery, includes data from FLEURS-derived speech, Mozilla Common Voice 25 Tajik, and augmented data from Muhtasham Tajik ASR. Each data split is provided in TSV format with an audio directory, and a SQLite version includes additional normalized fields.

Use Cases

• Train automatic speech recognition models based on the deduplicated Tajik speech data.
• Benchmark ASR system performance using the combined data from multiple public sources.
• Study language-specific acoustic features in Tajik using the provided audio and transcription pairs.
• Augment existing speech datasets using the normalized text and source metadata included in the SQLite version.

Strengths

• Data is deduplicated, which may reduce redundancy and improve model training efficiency.
• Combines speech data from at least three distinct sources, potentially increasing diversity.
• Provides data in both TSV and SQLite formats, with the SQLite version including extra fields like normalized_text.

Limitations

• Row count, total size, and audio file formats are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• License information is unknown, which could restrict commercial or research use.

Provenance

Source

FLEURS-derived speech data, Mozilla Common Voice 25 Tajik, Muhtasham Tajik ASR augmented data.

Collection Method

Assembled and deduplicated from the listed sources.

Freshness

Last updated 2026-05-29 12:01:47; freshness should be verified.