agentleFS
christos-antonochristos-antono/NPCTalk/.agents/skills/voice-input/SKILL.md

voice-input

Skill231 starsChanged 32 days agoMITView on GitHub

Let the player talk to NPCs out loud. Covers an always-on open mic with voice activity detection against the room's noise floor, push-to-talk and typing as fallbacks, ignoring the NPC's own voice coming out of the speakers, dropping coughs, clicks and filler words, fast speech-to-text (fal's Nemotron ASR, ~0.4 s, audio sent inline), and routing what was heard to the right NPC. Use when building or debugging mic input for a game, when the open mic triggers on noise or misses speech, when NPCs answer things the player didn't say, or when "nobody answers".

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep 300 ms from before the start, so the first syllable isn't cut.

    line 17

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Voice input

Three ways in, all ending in the same exchange: open mic (just talk), push-to-talk (hold N), type (B). Build push-to-talk first; it has no detection problems. Then the open mic, which is what makes it feel like talking to people.

Open mic (examples/gta-v/mod/OpenMic.cs)

  • 16 kHz mono capture in 30 ms blocks.
  • Noise floor: the quietest moments of the last 1.5 s. An utterance starts when the level stays SensitivityDb (12 dB) above that floor for 150 ms, and ends after SilenceMs (700 ms) below it. The floor adapts to the room and to the game's own sound.
  • Keep 300 ms from before the start, so the first syllable isn't cut.
  • Drop anything shorter than 450 ms (coughs, clicks) and anything over 12 s without a pause (music, a loud game).
  • Pause while an NPC talks, and for 600 ms after: their voice comes out of the speakers and would be transcribed as the player's. Also pause while push-to-talk is held, while typing, and while a reply is being written. Headphones make this much easier.
  • Drop non-speech after transcribing: one-word fillers ("uh", "um", "hmm", "yeah", "okay", "thanks"...) and empty transcripts are ignored (open mic ignored, not speech: in the log).

Who it goes to

Whoever the target lookup returns at that moment (skill make-npcs-talk, hook 1): within 5 m, or looked at within 12 m; the last partner sticks while close. If nobody's there, log it ("open mic -> nobody near") and show it on screen ("Listening... (to nobody near you)"), or the player thinks the mic is broken.

Speech-to-text

fal nvidia/nemotron-asr-multilingual/asr: send the WAV inline as a data URI (no upload step), ~0.4 s for a short sentence (0.38-0.80 s logged). Returns output or text. Whisper (any host) is the alternative; whatever you use, measure it per exchange.

Things that look like bugs

  • "Nobody answers" with people around: is the mod on? A toggle key that switches everything off looks exactly like a broken mic. Show the off state on screen. Then check the target rules (are they within 5 m or in front of the camera?).
  • The NPC answers the room. The open mic hears everything: a call, music, someone else talking. Give the player a mute key (M) and show the mic state on screen.
  • It triggers on the game. Raise SensitivityDb, or use headphones.
  • First word missing: the pre-roll is too short or the start threshold too slow.

Test without the game

npcvoice-check vad <voices folder> runs the detector on real voice clips mixed over quiet and loud noise, and reports what it caught and cut.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.