Skip to main content

​
Overview

Klaus uses Moonshine Voice for local, on-device speech-to-text transcription. No audio is sent to external APIs.

​
SpeechToText Class

The SpeechToText class provides WAV audio transcription using the Moonshine Voice library.

​
Constructor

from klaus.stt import SpeechToText
import klaus.config as config

stt = SpeechToText(settings=None)
​
settings
config.RuntimeSettings | None
default:"None"
Optional runtime settings. If None, reads from config.get_runtime_settings().

​
Methods

​
transcribe(wav_bytes: bytes) -> str

Transcribe WAV audio bytes to text.
​
wav_bytes
bytes
required
WAV-encoded audio data (mono or stereo, 16-bit PCM)
Returns: Transcribed text as a string. Example: 
from klaus.stt import SpeechToText

stt = SpeechToText()

with open("recording.wav", "rb") as f:
    wav_bytes = f.read()

transcript = stt.transcribe(wav_bytes)
print(transcript)

​
reload_settings(settings: config.RuntimeSettings | None = None) -> None

Reload settings and reinitialize the Moonshine transcriber.
​
settings
config.RuntimeSettings | None
default:"None"
New runtime settings. If None, reads from config.get_runtime_settings().

​
Configuration

STT settings are configured in ~/.klaus/config.toml: 
[stt]
backend = "moonshine"
moonshine_model = "base"  # or "tiny"
moonshine_language = "en"

​
Supported Models

  • tiny - Fastest, lower accuracy
  • base - Default, good balance of speed and accuracy

​
Supported Languages

  • en - English (default)
  • Additional languages may be supported depending on Moonshine Voice version

​
Implementation Details

  • Local processing: All transcription happens on-device. No network calls.
  • First-run download: The model is downloaded and compiled on first use (10-30 seconds).
  • Windows DLL preload: On Windows, moonshine.dll is preloaded before PyQt6 to avoid DLL conflicts.
  • Audio format: Accepts WAV files; automatically converts stereo to mono and normalizes to float32 in [-1, 1].

​
Source Reference

See klaus/stt.py for the full implementation.

Build docs developers (and LLMs) love

Get started for freeTalk to us