sprec(1)

okay bestie, this is sprec(1) from the General Commands Manual — literally the speech recognition command, and we’re gonna get you fluent in it. 💖✨

🎀 NAME 🎀

sprec — speech recognition command. that’s the whole name, no cap.

🦋 SYNOPSIS 🦋

the whole vibe is:

sprec [OPTIONS] < audio

🌸 DESCRIPTION 🌸

sprec reads audio from stdin and prints speech recognition results. that’s it, that’s the main character energy. 👑

💅 OPTIONS 💅

• --alts=NUM — specify the max number of alternative results. plural results, okay? the little buffet is real. 🍓
• -h, --help — print help and exit. your cute little escape hatch. 🫶
• --partial — also print intermediate results. yes, the in-progress tea. ☕
• --phrase-file=FILE — limit the vocab to phrases in a file. the phrases are split at whitespace unless quoted "like this". if the special phrase [unk] is included, it will be given for unrecognized words instead of the best match. so quotey quotes matter, bestie. 🙈
• --version — print the version and exit. cute. 💜

🧚 MODEL 🧚

the speech recognition model can be specified with the SPREC_MODEL environment variable — that’s the move. the default is /usr/share/vosk-models/small-en-us. sprec should work well with possibly all of the *-small and *-lgraph models from https://alphacephei.com/vosk/models. like, that’s the “possibly all” vibe, bestie. 🌈

🔥 AUDIO 🔥

the correct audio format depends on the model, but possibly all the Vosk models like WAVE 16000Hz Mono, which you can record with:

arecord -q -fS16_LE -c1 -r16000

so vibe check your format first, okay? 💀

⭐ EXAMPLES ⭐

print realtime speech recognition results:

arecord -q -fS16_LE -c1 -r16000 -D sysdefault:CARD=Microphone | spec

see the sprec_type script for an example that types out the results (https://git.sr.ht/~geb/sprec/examples/sprec_type). that’s the typing-out-results demo, iykyk. 🎀

👑 AUTHOR 👑

John Gebbie. icon. 💅

---

🎀 final footer energy: source/version sprec(1), dated 2026-10-14. that’s the cute sign-off, bestie, bye! 💖🫶