GigaSpeech ASR Clean: Filtered Speech Recognition Samples

A filtered dataset for automatic speech recognition (ASR) created by OpenSpeechHub. The dataset has been cleaned by removing samples with fewer than three words, repetitive tokens, or chat token leaks. It was last updated on March 31, 2026.

Use Cases

• Training speech recognition models based on filtered audio samples.
• Benchmarking ASR model robustness against short or repetitive utterances based on the described cleaning criteria.
• Studying the impact of data filtering on ASR performance based on the removal of low-quality samples.

Strengths

• The dataset has been filtered to remove low-quality samples, specifically those with fewer than three words.
• Additional cleaning steps have removed samples containing repetitive tokens or chat token leaks.
• The dataset was last updated on March 31, 2026, suggesting recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

OpenSpeechHub

Collection Method

Filtered from a larger ASR dataset; specific method not detailed.

Time Range

null

Freshness

Last updated 2026-03-31 21:37:48.

Geography

null