Local Speech-to-Text with LFM2.5-Audio-1.5B and llama.cpp

Melvin Vivas · X video post · 2026-09-08 · 0:16 · 600 views · Open on X

Topics: LLMOps, Deployment & Monitoring, LLM Fundamentals · Level: intermediate

Summary

This is a 16-second demo with no speech. It shows a speech-to-text (audio-to-text) CLI that runs entirely on a laptop. The tool serves Liquid AI's LFM2.5-Audio-1.5B model with llama.cpp. The quoted post links to the open-source code in the Liquid4All cookbook, which was recently updated to use the LFM2.5 model family.

Key points

• LFM2.5-Audio-1.5B is a small (1.5B-parameter) audio model from Liquid AI's LFM2.5 family that can do speech-to-text.
• llama.cpp can serve the model on a laptop, so transcription runs 100% locally with no cloud API.
• The demo is a command-line tool that transcribes any audio input you give it.
• The example code is in the Liquid4All cookbook repo under examples/audio-transcription-cli.
• The demo was recently switched to the LFM2.5 family, replacing an earlier model version.
• Small audio models plus llama.cpp are a practical way to get private, offline, low-cost transcription.

Resources mentioned

• Liquid4All Cookbook – audio-transcription-cli example · repo · github.com · free
  Open-source example code for a fully local audio-to-text CLI that serves LFM2.5-Audio-1.5B with llama.cpp.
• LFM2.5-Audio-1.5B · tool · huggingface.co · free
  Liquid AI's 1.5B-parameter audio model from the LFM2.5 family, used here for speech-to-text.
• llama.cpp · repo · github.com · free
  An open-source C/C++ engine for running GGUF models locally. Its llama-server command provides an OpenAI-compatible HTTP server.
  Also in: Running LLMs Locally Without an Expensive Rig (Melvin Vivas on X · notes), llama.cpp / Llama-macOS v0.5.0 release (Melvin Vivas on X · notes), Run llama.cpp GGUF Checkpoints in Hugging Face Transformers (Melvin Vivas on X · notes), llama.cpp v0.4.1 release announcement (Melvin Vivas on X · notes) and 28 more
• Liquid AI (@liquidai) on X · website · x.com · free
  An AI company that builds efficient foundation models. The quoted post shows its PII handling working on Japanese text.
  Also in: Zero-Shot Prompt Routing with Liquid AI's LFM 2.5-Encoder-350M (Melvin Vivas on X · notes), Liquid AI LFM 2.5 Encoder: a CPU-friendly encoder model (Melvin Vivas on X · notes), Fine-tuning Liquid AI LFM2/LFM2.5 MoE models with the new Halo framework (Melvin Vivas on X · notes), Liquid AI's LFM2-Longevity models for aging-data analysis (Melvin Vivas on X · notes) and 19 more

Try this

• Clone the Liquid4All cookbook and run the audio-transcription-cli example.
• Serve LFM2.5-Audio-1.5B with llama.cpp on your laptop and transcribe your own audio files.
• Build a fully local, private audio-to-text transcription CLI using LFM2.5-Audio-1.5B and llama.cpp.