Skip to main content
The MicTranscriber class extends Transcriber with automatic microphone audio capture, making it easy to transcribe live speech from the system’s default microphone.

​
Class Definition

from moonshine_voice import MicTranscriber, ModelArch

mic_transcriber = MicTranscriber(
    model_path: str,
    model_arch: ModelArch = ModelArch.BASE,
    update_interval: float = 0.5,
    options: dict = None,
    device: int = None,
    sample_rate: int = 16000,
    channels: int = 1,
    blocksize: int = 1024
)

​
Constructor Parameters

Inherits all Transcriber parameters plus:
​
device
int
default:"None"
Audio input device index. None uses the system default microphone.Use sounddevice.query_devices() to list available devices.
​
sample_rate
int
default:"16000"
Microphone sample rate in Hz. 16000 is optimal for Moonshine models.
​
channels
int
default:"1"
Number of audio channels. Must be 1 (mono) for Moonshine.
​
blocksize
int
default:"1024"
Audio buffer size in samples. Smaller values reduce latency but may cause audio dropouts.

​
Methods

Inherits all Transcriber methods. The start() and stop() methods automatically handle microphone capture:

​
start

Begin microphone capture and transcription. 
mic_transcriber.start()
Opens the audio input device and starts feeding audio to the transcriber. Audio is captured continuously until stop() is called.

​
stop

Stop microphone capture and end the session. 
mic_transcriber.stop()
Closes the audio device and finalizes the transcript.

​
Example: Basic Microphone Transcription

from moonshine_voice import MicTranscriber, ModelArch, TranscriptEventListener

class PrintListener(TranscriptEventListener):
    def on_line_text_changed(self, event):
        # Print updates as they come
        print(f"\r{event.line.text}", end="", flush=True)
    
    def on_line_completed(self, event):
        # Print final version on new line
        print(f"\n{event.line.text}")

mic_transcriber = MicTranscriber(
    model_path="/path/to/models",
    model_arch=ModelArch.TINY_STREAMING
)

mic_transcriber.add_listener(PrintListener())

print("Listening... Press Ctrl+C to stop")
try:
    mic_transcriber.start()
    while True:
        pass  # Keep running
except KeyboardInterrupt:
    mic_transcriber.stop()
    print("\nStopped")

​
Example: Selecting Audio Device

import sounddevice as sd
from moonshine_voice import MicTranscriber

# List available devices
print(sd.query_devices())

# Use device index 2
mic_transcriber = MicTranscriber(
    model_path="/path/to/models",
    device=2,
    sample_rate=44100  # Device's native sample rate
)

​
Command-Line Usage

You can run the MicTranscriber directly from the command line: 
python -m moonshine_voice.mic_transcriber --language en
  • --language - Language code (en, es, ar, ja, ko, zh, uk, vi)
  • --model-arch - Model architecture (0-5)
  • --model-path - Path to model files
  • --device - Audio device index
  • --sample-rate - Sample rate (default: 16000)
  • --update-interval - Update interval in seconds
  • --options - JSON string of advanced options

​
Audio Processing Pipeline

The MicTranscriber handles the complete audio pipeline:
  1. Capture - sounddevice captures audio from microphone
  2. Buffer - Audio chunks are buffered
  3. Feed - Chunks are fed to the transcriber via add_audio()
  4. Process - VAD, segmentation, and transcription happen automatically
  5. Events - Listeners receive updates as speech is recognized

​
Configuration Tips

Use ModelArch.TINY_STREAMING for lowest latency on resource-constrained devices.
Set channels=1 (mono). Moonshine doesn’t support stereo input.
If you experience audio dropouts, increase blocksize to 2048 or 4096.

​
Permissions

Microphone access requires user permission on most platforms:
  • macOS: Add microphone usage description to Info.plist
  • Windows: Ensure microphone privacy settings allow access
  • Linux: User must be in audio group

​
Troubleshooting

  1. Check microphone is working with other apps
  2. Verify correct device index with sounddevice.query_devices()
  3. Check VAD threshold - try lowering it: 
    options = {"vad_threshold": "0.3"}
  4. Enable debug logging: 
    options = {"log_output_text": "true"}
  1. Increase blocksize: blocksize=2048
  2. Close other audio applications
  3. Check system CPU usage
  4. Try a lower sample rate: sample_rate=8000

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us