AI Engineering

Jan 6, 2026 · Artificial Intelligence

1.5B‑Parameter Model Enables Offline Real‑Time Speech Transcription

Liquid AI’s new 1.5 B‑parameter LFM2‑Audio model delivers high‑quality offline, real‑time speech‑to‑text, text‑to‑speech, and multimodal dialogue on local devices, using a 1.2 B language backbone, a FastConformer encoder, and supports two generation strategies, with benchmark scores surpassing larger rivals.