YodaLingua-Swedish: 112 Hours of Swedish Speech for TTS and ASR

Swedish speech data containing 43,048 audio-transcription pairs totaling 112 hours of audio from 1,946 distinct speakers. The YodaLingua-Swedish dataset, created by Thomcles, is designed for training text-to-speech and automatic speech recognition systems. It was last updated on the Hugging Face platform in January 2026.

Use Cases

• Train text-to-speech (TTS) systems based on the described high-quality, clean audio-text pairs.
• Develop automatic speech recognition (ASR) models based on the well-aligned audio-transcription pairs.
• Fine-tune voice cloning or speech synthesis models based on the data from 1,946 distinct speakers.
• Benchmark multilingual speech models using the Swedish portion of the described YodaLingua collection.

Strengths

• Contains 43,048 individual audio clips, providing a substantial volume of data.
• Offers 112 hours of total audio duration for model training.
• Includes speech from 1,946 distinct speakers, suggesting speaker diversity.
• Described as 'high-quality' and containing 'clean, well-aligned audio–text pairs' for core speech tasks.

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

Hugging Face (dataset uploaded by Thomcles).

Collection Method

Likely a curated collection of speech recordings, but specific gathering method is not detailed.

Freshness

Last updated 2026-01-25 19:56:58; freshness should be verified.

Geography

Sweden (implied by language focus).