Documentation›Models, images, and voice
33 Models and media

Configure Voice, Dictation, and Calls

Set up speech recognition, voices, dictation, and live calls.

Local Waifu 1.8.0 macOS Windows
plain English first

Follow the bold button names and numbered steps. Technical names are included only when you need to recognize a model or file inside the app.

Local Waifu has three separate voice functions:

  1. Dictation: speech becomes editable text in the composer; it is not sent automatically.
  2. Voice output: the companion reads a response aloud.
  3. Call: a continuous speech-recognition → chat → speech loop.

Four more voice features, covered in Chapter 34, are voice messages, hands-free start, bedtime stories, and songs.

Install speech and voice models

A full call needs:

  • speech recognition (STT/Whisper) to hear you;
  • a voice model (TTS/Supertonic or another configured route) to speak.

Missing STT blocks the full call. Missing high-quality TTS may fall back to an older voice, but pronunciation can be worse.

Open Settings → Models → Voice Model, install Speech recognition and Voice model (Set up voice installs both), and optionally install Natural Voice. If the speech model is English-only, the app tells you to remove it and download the multilingual one. The screen reports current package sizes and disk use.

Choose a voice

  1. Open Settings → Models → Voice Model.
  2. Preview the default, female, or male voices with the play button.
  3. Select a voice for the active character.
  4. Turn on Natural voice for the warmer, more human local voice when installed. It is a bigger download and answers a little slower (about 3 seconds to first sound), so it is off by default. On macOS it runs on Apple Silicon in the Direct edition only and needs 16 GB of memory. The tab says when your Mac cannot run it.
  5. Use Test her voice, type a sentence, and select Speak.

Language support depends on the voice engine and the platform.

  • Both platforms: the standard voice covers English, Polish, German, Spanish, Japanese, and Korean.
  • Windows: Chinese needs Natural voice turned on and ready. Natural voice also supports Korean.
  • macOS: the voice model has no Chinese at all, so Chinese falls back to the older English-accented voice. With Natural voice, Korean falls back to the default voice.

Under Voice language you choose the language she speaks, even when the app is in another one. It does not change how she understands you.

Add your own voice sample

Goal

Create a character-specific voice reference.

Before you start

  • Record 10 to 30 seconds of clean speech.
  • Use one speaker in a quiet room.
  • Record in the language/accent you want to preserve.
  • File size must be under 25 MB.

Steps

  1. Open Settings → Models → Voice Model.
  2. Under Your own voice, select Add a voice from a file.
  3. Choose the audio file.
  4. Enter a short name, up to 60 characters.
  5. Wait for local processing and transcription.
  6. Preview the result and select it.

Expected result

The active character uses the new reference and Natural Voice is enabled when ready.

If something goes wrong

Use a shorter, cleaner sample. A learned English sample keeps its English accent in other languages; record again in the target language for better pronunciation.

Tune how she speaks

Available style presets include Auto, Soft & warm, Balanced, Bright & playful, Calm & even, and Confident & teasing.

Adjust:

  • speaking pace;
  • pitch;
  • expressiveness;
  • Let her mood colour her voice.

Use small changes and test after each one. Auto follows the character personality; manual presets keep your chosen delivery style.

Configure a call

Goal

Select audio devices and tune turn-taking.

Steps

  1. Open Settings → Models → Voice Model and find The call.
  2. Select Microphone and Speaker, or keep System default.
  3. Select Test microphone and verify activity.
  4. Adjust How long she waits. A shorter value feels faster but can cut you off mid-thought.
  5. Enable I’m wearing headphones only when appropriate; speakers can feed her voice back into the microphone and cause self-interruption.
  6. Leave cloud recognition and cloud voice off for fully local processing, or enable them intentionally with your provider key.
  7. Start a call with the phone button (Talk to her) or Command/Ctrl+Shift+C.
  8. Optionally turn on Hands-free to start a call by saying her name. Chapter 34 covers it.

Expected result

Your speech is transcribed, answered by the active model, and spoken through the selected voice and speaker.

If something goes wrong

  • Grant microphone permission in the operating system.
  • Choose the correct input device.
  • Install STT if the call cannot hear you.
  • Install/configure TTS if text replies appear but no voice plays.
  • Increase the waiting time if you are interrupted too early.