Back to News
ProductAI Understanding briefing

Klang raises €1.32 million and launches open‑source Swedish speech‑to‑text model

Swedish‑focused conversation AI startup Klang announced a €1.32 million funding round and released its first open‑source speech‑to‑text model, Pianissimo, claiming speed and accuracy comparable to the leading national‑library model.

4 min readRead the linked source
Source-provided image accompanying Klang raises €1.32 million and launches open‑source Swedish speech‑to‑text model
Source referenceSource recorded
Publisher
eu-startups.com
Source link
eu-startups.comhttps://www.eu-startups.com/2026/09/helsingborgs-conversation-ai-startup-klang-raises-e1-32-million-and-releases-open-speech-to-text-model-for-swedish/
Source type
Linked source — primary-source status has not been established.
ContextUnderstand this in 60 seconds

Start here

Key terms

Inference
The runtime phase where a trained model generates predictions or outputs.
Latency
The time between sending a request and receiving the model's output.
Test yourselfAI Models Explained Quiz

What happened

Klang secured €1.32 million (SEK 15 million) in fresh capital at a €13.2 million valuation and published an open‑source speech‑to‑text model for Swedish, named Pianissimo.

According to EU‑Startups, Klang—a Helsingborg‑based conversation AI company founded in 2023—announced a €1.32 million financing round led by entrepreneur and tech investor Johan Lenander, with participation from Emil Sjödin (founder of Refined) and Daniel Gadd (co‑founder of Position Green). The round was described as a fresh capital injection to support continued AI development and international expansion, with plans to hire roughly ten staff in research, development, and marketing.

In the same announcement, Klang released its first open‑source speech‑to‑text model for Swedish, called Pianissimo. The model is built on Nvidia’s Parakeet architecture, which the company says it has optimized for Swedish and its dialects. Klang claims the model can run on a standard laptop, making it accessible for developers without specialized hardware.

Klang reported internal testing that shows Pianissimo achieving accuracy comparable to the KB Whisper Large model—identified by the National Library of Sweden as the best open Swedish model—while delivering speeds about 60 times faster. The company positioned this speed advantage as a key enabler for real‑time transcription services.

Source details: eu-startups.com ↗

Why it matters

The release adds a high‑performance, locally runnable Swedish transcription model to the open‑source ecosystem, potentially lowering barriers for developers building real‑time conversation tools in a language that has few fast, accurate options. By basing the model on Nvidia’s Parakeet and claiming roughly 60× faster than the best publicly available Swedish model, Klang could enable new low‑cost services on consumer hardware, expanding the reach of AI‑driven transcription in Europe and beyond.

Swedish speech‑to‑text tools are relatively scarce, and existing solutions often require cloud resources or incur that hampers real‑time applications. By providing a model that runs locally and claims high speed, Klang addresses a practical gap for enterprises and developers needing on‑device processing for privacy or low‑bandwidth scenarios.

Open‑source releases encourage community contributions, benchmarking, and rapid iteration. If independent researchers validate Klang’s performance claims, Pianissimo could become a reference implementation for other Nordic language models, fostering a broader ecosystem of multilingual, efficient transcription tools.

The funding round signals investor confidence in niche language AI solutions and may catalyze further investment in European‑focused AI startups, potentially accelerating the development of locally tailored models that respect regional data‑protection standards.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Model Parameter Size:8B Parameters
VRAM Required5.5 GBGPU memory footprint
Target HardwareMacBook / Single GPUDeployment tier
Privacy100% Air-GappedLocal device capability
Core takeaway: Small, quantized models (3B–8B) now run directly inside smartphones and laptops with complete data privacy, while mammoth 400B+ models remain the domain of datacenter clusters.
Interactive Concept Check+10 Points
AI Models Explained Quiz

In AI, what are a model's "parameters"?

What to watch next

Future developments to monitor include Klang’s rollout of the model in the Nordics, Benelux, DACH, and Japan, as well as any performance benchmarks released by independent parties. Adoption by third‑party developers, especially in privacy‑sensitive sectors, will indicate whether the speed claims hold up in diverse environments. Additionally, the impact of the new funding on Klang’s product roadmap and any subsequent model releases will be key signals of the company’s growth trajectory.

Verification of Pianissimo’s speed and accuracy by third‑party benchmarks, especially in varied hardware environments.

Adoption metrics such as the number of GitHub forks, community contributions, and integration into commercial products within the next six months.

Klang’s expansion plans into the Nordics, Benelux, DACH, and Japan, including any localized versions of the model or new language offerings.

Potential follow‑on funding rounds or strategic partnerships that could broaden the model’s distribution or integrate it with larger AI platforms.

Related guides & quizzes

AI Models ExplainedTransformersFuture of AIAI EthicsTest what you know — try a free AI quizLook up an AI term in our glossaryFollow the AI model release tracker
Found this useful?