Aug 24, 2026 · 4 min read

LM Studio voice chat from your phone

Voice Chat with Local AI

Typing is great until your thumbs file a complaint. Voice chat is my peace treaty with them. And my escape hatch for apparent laziness.

What "voice chat" means here

In LM Mini, voice is a layer on top of the same local stack you already use.

You speak and the app turns that into text. The model replies and the reply can be read back to you. Optionally it keeps listening so the conversation flows like a call, not like a dictation contest.

If that sounds simple, good. The hard part was making it feel normal on a phone.

Why I added it

Feature requests. Always feature requests.

People wanted to cook and ask questions, or they would rather work their mouths instead of their thumbs. Giving them the benefit of the doubt: walk and brainstorm, sit in a parked car and dump a messy thought before it evaporates.

Also, Personas got more fun when they could sound like themselves. A calm tutor should not sound like a robot auctioneer. A chaotic idea generator can sound slightly unhinged. We call this Art.

The pieces (plain English)

You do not need to memorize every setting on day one. Mentally, there are a few knobs:

  • Speech-to-text (STT). Your voice becomes the prompt. Options include on-device / system speech recognition and Whisper-style setups depending on platform. Goal: you talk, text appears, model gets to work.
  • Text-to-speech (TTS). The reply gets a voice. This is where Kokoro shines for people who want nicer, more characterful speech, including voices tied to Personas.
  • Continuous conversation. Say something, hear the answer, keep going without stabbing the mic button like it owes you money.
  • Call Mode. Lean into the phone-call feeling. Longer back and forth. Fewer "wait, is it listening?" moments.

Group Chat can do group voice too, which is delightful and slightly chaotic. Headphones help. Your dignity may not.

Voice + Personas = actually fun

This is the part I did not expect you to love as much as I do.

Give a Persona a voice. Suddenly your writing coach sounds patient. Your code reviewer sounds strict. Your study buddy sounds like someone who will not let you skip the quiz.

Same prompt library. Different presence. It turns "AI chat" into "I am talking to a character I designed," which is either productive or a new hobby. Both are valid.

Local still matters with the mic on

Microphones make people nervous, for good reasons.

Voice in LM Mini is still aiming at the same privacy story: on-device where you run on-device, local network when you use LM Studio or Ollama at home, encrypted remote through LM Mini Connect (Windows/Linux) or Home when you are away and on Pro. Turning on the mic should not quietly redefine "local" as "lol never mind."

Practical tips from someone who has embarrassed himself enough for science:

  • Check which STT path you are using if you are picky about where audio processing happens.
  • Do not voice-chat secrets on speakerphone in a cafe and then blame the app. Physics exists.
  • Start with short sessions. Continuous mode is powerful. It is also how you accidentally ask your pasta recipe to refactor your codebase.

If Kokoro on a Windows PC fails to download in Connect 1.5.0, update to 1.5.1 — or extract the archive with 7-Zip and use Use folder.

A simple way to try it

  • Pick a model you already like (on-device for privacy and simplicity, or your home LM Studio / Ollama box for bigger brains).
  • Optional: open a Persona and assign a Kokoro voice.
  • Enable voice / tap the mic in chat.
  • Ask something you would normally type: "Explain this like I am tired" works shockingly well.
  • If you want flow, turn on continuous conversation or Call Mode and go for a walk. Watch for traffic. The model cannot catch a bus for you.

If replies sound weird, tweak voice speed and Persona instructions before you blame the universe. Half of "bad voice AI" is just a Persona that thinks it is in a Shakespeare play.

When voice is better than typing

  • Brain dumps and brainstorming
  • Studying out loud
  • Hands busy, mind free
  • Accessibility when typing is annoying or painful
  • Making Personas feel alive without reading a wall of text

When typing is better: code-heavy edits, long pastes, quiet offices, and any moment where talking to your phone would make strangers form a support group.

The bottom line

Voice chat is not a gimmick glued on top of LM Mini. It is another door into the same private room: local models, your Personas, your pace.

Try one conversation with your thumbs unemployed. If you like it, assign voices to your cast and never look back. If you hate it, the keyboard is still there. No judgment. Only mild judgment.

Next up: which models actually run well on your phone — including the occasional rant about RAM I have been saving. → Which Models Run Well on Your Phone?

Get LM Mini: lmmini.com

LM Studio voice chat from your phone

If you search for lm studio voice chat, you usually want the model to stay on your computer (LM Studio) while your phone handles the mic and speaker. That is exactly how LM Mini voice works: STT on the phone, chat against your LM Studio server (typically :1234), TTS back so you can listen hands-free.

Same pattern for Ollama. Pair a Persona with a Kokoro voice if you want a consistent character. Keep the PC awake; use Wi‑Fi first, then LM Mini Home / VPN when you leave the house. Accurate claim: LM Mini is the phone client — LM Studio still runs the model.