Dutch Speech Audio with Transcripts from Common Voice

Common Voice NL 8 Processed is a Dutch-language subset of Mozilla's crowdsourced speech corpus. The dataset was uploaded to Hugging Face by user RuudVelo in February 2022, indicating processing of the eighth version of the Dutch Common Voice data. It contains audio clips paired with text transcriptions for speech technology development.

Use Cases

• Train an ASR model to predict text transcriptions from Dutch audio clips.
• Fine-tune a speech encoder using the audio data and corresponding text labels for representation learning.
• Analyze speaker demographics or regional accents using metadata like `region` or `age` if present in the source Common Voice data.
• Benchmark model performance on Dutch speech using standardized train/validation/test splits typical of Common Voice releases.
• Create a pronunciation dictionary or language model for Dutch using the collected text transcriptions.

Strengths

• Derived from Mozilla Common Voice, a structured crowdsourcing project with validation mechanisms.
• Focuses on Dutch, a mid-resource language for speech technology.

Limitations

• Exact size, row count, and audio duration are unknown from available metadata.
• Potential for label noise inherent in crowdsourced transcription tasks.
• Data is from 2022 or earlier, which may not reflect recent language use.

Provenance

Source

Mozilla Common Voice project.

Collection Method

Crowdsourced audio recordings and transcriptions.

Freshness

Last updated on the platform in February 2022; update frequency is unknown.

Geography

Primarily Dutch-speaking regions, inferred from language focus.