sunfounder_stt

Speech-to-Text brick — push-to-talk recording via ALSA, transcription via Whisper.

Engines

All backends accessed through a single STT class — switch by changing the type= parameter:

from sunfounder_stt import STT

type=	Engine	Online	Notes
local_fast (default)	faster-whisper CTranslate2	No	int8 CPU, auto-downloads ~72 MB
local_standard	whisper.cpp GGML	No	needs model= path, e.g. /app/models/ggml-tiny.bin (42 MB)
online	OpenAI Whisper API	Yes	needs API_KEY and internet

Usage

local_fast — faster-whisper (default, recommended)

from sunfounder_stt import STT

stt = STT(type="local_fast", language="zh")

# Button press
stt.start_listening()
# ... user speaks ...

# Button release
stt.stop_listening()
text = stt.get_result()  # blocks until done, returns transcribed text

local_standard — whisper.cpp GGML

stt = STT(type="local_standard", model="/app/models/ggml-tiny.bin", language="zh")
# same API: start_listening() / stop_listening() / get_result()

online — OpenAI Whisper API

STT.API_KEY = "sk-..."
stt = STT(type="online", language="zh")
# same API

API Reference

Method	Description
start_listening()	Spawn arecord -D hw:0,2 and begin capturing PCM
stop_listening()	Terminate capture, save WAV, trigger transcription
get_result(timeout=30) -> str	Block until transcription completes (empty string on timeout)
is_ready() -> bool	Has transcription completed? (non-blocking)
reset()	Discard pending recording and result

Constructor parameters

Parameter	Default	Description
type	"local_fast"	Backend: "local_fast", "local_standard", or "online"
model	None	Model path for local backends ("tiny" if omitted)
language	"en"	Language code, e.g. "zh", "en", "auto"
samplerate	16000	Audio sample rate in Hz

Audio Capture

Audio is captured via ALSA direct: arecord -D hw:0,2 (S16_LE, 16000 Hz, mono). PCM saved as WAV to /app/audio_output/stt_last.wav.

The Qualcomm Codec capture mixer is configured at container startup by setup_audio_output() in robot_shield.

Host pre-requisite:

systemctl --user stop pipewire pipewire-pulse wireplumber

Dependencies

• faster-whisper (for local_fast; default)
• pywhispercpp (for local_standard)
• requests (for online)
• arecord from alsa-utils (system package)