YodaLingua Croatian: 11 Hours of Speech for Text-to-Speech and ASR Models

YodaLingua-Croatian is a speech dataset by Thomcles containing 5,655 audio-transcription pairs. The collection totals 11 hours of Croatian speech from 230 distinct speakers, last updated in January 2026. It is designed for training text-to-speech and automatic speech recognition systems.

Use Cases

• Train text-to-speech (TTS) systems based on the high-quality, well-aligned audio-text pairs.
• Develop automatic speech recognition (ASR) models based on the clean Croatian speech recordings.
• Fine-tune voice activity detection models based on the mono audio format and speaker diversity.
• Build multilingual speech applications based on the dataset's inclusion in a larger multilingual collection.

Strengths

• 5,655 audio clips provide a substantial number of training examples.
• 11 hours of total audio duration offers significant speech data.
• 230 distinct speakers suggest potential diversity in vocal characteristics.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Thomcles on Hugging Face.

Collection Method

Likely a curated collection of speech recordings aligned with transcriptions.

Time Range

null

Freshness

Last updated 2026-01-25 17:44:14; freshness should be verified.

Geography

Croatian language focus suggests potential regional coverage of Croatia or Croatian-speaking areas.