Files

64 lines
1.7 KiB
INI

# voice_linux runtime configuration
# Hotkey format supports Ctrl, Alt, Shift, Super + final key symbol
# Example: Ctrl+Alt+V
model_path=./models/ggml-medium.en.bin
use_gpu=1
audio_device=alsa_input.usb-Shure_Inc_Shure_MV6_MV6_5-ed2d37fa14d3e65d8d477cacf910d93c-01.mono-fallback
sample_rate=16000
language=en
hotkey=Ctrl+Alt+V
type_delay_us=8000
max_record_seconds=30
# always-on VAD controls
always_on=0
# legacy window size (unused by new VAD pipeline, kept for compatibility)
always_on_window_ms=1400
# live level threshold (0.0 - 1.0)
vad_peak_threshold=0.0460
# level must stay above threshold this long before speech starts
vad_trigger_ms=153
# level must stay below threshold this long before speech ends
vad_release_ms=300
# include this much pre-trigger audio in transcript segment
vad_preroll_ms=500
# discard segments shorter than this
vad_min_speech_ms=692
# hard cap for one speech segment
vad_max_speech_ms=15000
autotype_enabled=1
filter_bracketed_tags=0
# whisper decode controls
whisper_sampling_strategy=0
whisper_n_max_text_ctx=16384
whisper_offset_ms=0
whisper_duration_ms=0
whisper_translate=0
whisper_detect_language=0
whisper_no_context=1
whisper_no_timestamps=1
whisper_single_segment=1
whisper_token_timestamps=1
whisper_thold_pt=0.0100
whisper_thold_ptsum=0.0100
whisper_max_len=0
whisper_split_on_word=0
whisper_max_tokens=0
whisper_audio_ctx=0
whisper_tdrz_enable=0
whisper_suppress_blank=1
whisper_suppress_nst=1
whisper_temperature=0.0000
whisper_max_initial_ts=1.0000
whisper_length_penalty=-1.0000
whisper_temperature_inc=0.2000
whisper_entropy_thold=2.4000
whisper_logprob_thold=-1.0000
whisper_no_speech_thold=0.6000
whisper_greedy_best_of=1
whisper_beam_size=5
whisper_beam_patience=-1.0000