ovos-microphone-plugin-pyaudio

This microphone plugin for OVOS runs on Linux, macOS, and Windows through PortAudio via pyaudio. It does not depend on speech_recognition. It calls the PyAudio API directly and maps sample_width to the matching PyAudio format for int16, int32, and float32 capture.

Platform: Linux, macOS, Windows (requires the libportaudio2 system library).

Source: PyAudioMicrophone in ovos_microphone_plugin_pyaudio/__init__.py

Entry point: opm.microphone = ovos-microphone-plugin-pyaudio

---

Configuration

Field	Type	Default	Description
device	str|int|None	config hierarchy / "default"	Device name, integer index, or "default"
period_size	int	1024	Frames per stream.read() call
timeout	float	5.0	Seconds to block in read_chunk() before returning None
multiplier	float	1.0	Gain multiplier (skipped for float32, requires audioop)
float32_output	bool	False	Use paFloat32, required by ggwave
muted	bool	False	Enqueue silence instead of real audio
queue_maxsize	int	8	Max buffered chunks, oldest evicted when full
sample_rate	int	16000	Sample rate in Hz (base class)
sample_width	int	2	Bytes per sample (base class)
sample_channels	int	1	Channels (base class)
chunk_size	int	4096	Output chunk size in bytes (base class)

sample_width → PyAudio format mapping

sample_width	PyAudio format
1	paInt8
2	paInt16
3	paInt24
4	paInt32
float32_output=True	paFloat32 (overrides width)

---

float32 output (ggwave)

{
  "listener": {
    "microphone": {
      "module": "ovos-microphone-plugin-pyaudio",
      "ovos-microphone-plugin-pyaudio": {
        "sample_rate": 48000,
        "sample_width": 4,
        "sample_channels": 1,
        "float32_output": true
      }
    }
  }
}

---

Device selection

find_input_device(name) tries, in order: integer passthrough, exact name, substring, then regex. list_input_devices() returns (index, info_dict) pairs for all input-capable devices.

Device resolution order

1. listener.microphone.ovos-microphone-plugin-pyaudio.device
2. listener.device
3. "default"

---

Known limitations

• No resampling or channel conversion
• No stream-open fallback chain
• multiplier gain requires audioop (removed from Python 3.13 stdlib)