Speechmatics · 4× · naposledy 26. 9. 2026
48
Momentum
Linden is a Speech-to-Text model developed by Speechmatics, specifically designed for real-time voice agents rather than general transcription. It forms the foundation of the "Agent STT" product and recognizes speech in over 55 languages, including accents, dialects, and alphanumeric information such as account or card numbers. The model finalizes spoken segments in under 350 milliseconds according to internal tests and achieved a semantic error rate of 1.05% in an independent Pipecat Benchmark with a median finalization time of 369ms. Linden is available via the Speechmatics API, LiveKit Inference, and Pipecat.
Vývoj momenta
29.06.27.09.