Speech recognition
Pianissimo: The best Danish speech-to-text model
Fast, accurate and open: Danish speech recognition.
November 3, 2026 · Mattias Fält · 3 min read
Klang Research
Advancing voice technology for the Nordics and beyond. Open foundations for conversational understanding.
Speech recognition
Why benchmarking is hard: a Danish ASR study
Sentence overlap, training data and Danish speech recognition benchmarks.
November 3, 2026 · Mattias Fält · 7 min read
Speech recognition
How we trained Pianissimo
How we built the training data, refined the model and fixed unexpected problems with long recordings.
September 23, 2026 · Mattias Fält · 15 min read
Datasets
We asked people to break Klang: a new benchmark for Swedish dialects
A public challenge became an open Swedish speech benchmark, with recordings from across Sweden.
September 14, 2026 · Mattias Fält · 6 min read
Voice
Voice recognition in Klang
How voices become embeddings, how much one voice varies compared to another, and why we had to warp the embedding space. Part one of two.
August 11, 2026 · Mattias Fält · 5 min read
Voice
Unlikely similar voices
How many clips do we need to identify a voice, and how strict can we make the cutoff before recognition suffers? Part two of two.
August 11, 2026 · Mattias Fält · 5 min read
Publications
DateCategoryTitle
- Speech recognitionPianissimo: The best Danish speech-to-text model
- LanguageThe Nordic token tax
- Speech recognitionIntroducing Klang Pianissimo
- Speech recognitionWhy benchmarking is hard: a Danish ASR study
- Speech recognitionHow we trained Pianissimo
- DatasetsWe asked people to break Klang: a new benchmark for Swedish dialects
- VoiceVoice recognition in Klang
- VoiceUnlikely similar voices