FPT FOSD: Vietnamese Speech Dataset with 25.9k Samples

25,900 audio samples totaling 100 hours of Vietnamese speech data, originally released by FPT Corporation in 2018. This is an unofficial mirror hosted by 'doof-ferb' after the official link became inactive. The data has been pre-processed to remove non-sense strings and four files missing transcriptions.

Use Cases

• Train automatic speech recognition (ASR) models based on the 100 hours of Vietnamese speech.
• Benchmark speech-to-text system performance using the 25.9k audio-transcription pairs.
• Fine-tune pre-trained multilingual speech models on Vietnamese data.
• Analyze speech characteristics and transcription accuracy in a Vietnamese corpus.

Strengths

• Contains 100 hours of speech data, providing substantial material for model training.
• Includes 25,900 individual audio samples, offering a high volume of data points.
• Has undergone pre-processing to remove non-sense strings and files with missing transcriptions.

Limitations

• The original source link is dead, raising questions about official support and updates.
• Column-level documentation and sample data are unavailable, making detailed inspection difficult.
• The description notes a need to 'check misspelling', suggesting potential quality issues in the transcriptions.

Provenance

Source

FPT Corporation

Collection Method

Released publicly.

Freshness

Last updated on the platform 2024-02-10 11:23:24; the underlying data is from 2018.