voice-references

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: none for the install itself: no account, no API key, no paid service, and nothing else to install first. For using the skill afterwards as described, you need a fal account and API key (environment variable FAL_KEY) for fal speech-to-text transcription, the npctalk CLI and ffmpeg on PATH (both live outside this skill folder and are not installed by this prompt), a recording you have the rights to, and, to upload and use the clips, your own Fish Audio voice server on Modal (its npcvoice-models volume, read at server start; the modal CLI does the upload). SKILL.md and the README name no price and no paid plan for fal or Modal. Never put any key in a file. Install "voice-references" for me. It is a single SKILL.md playbook for making the reference clips a zero-shot voice cloner needs for each character (Fish Audio Drama 3 or S2 Pro, MiniMax, ElevenLabs-style cloning) from recordings you have the rights to: split into lines, pick clean 6-11 s runs, transcribe them, drop screams and noise, build a ~25 s clone clip plus single reference clips and mood clips (angry, drunk), check them, and upload them to the voice server. Repository: https://github.com/christos-antono/NPCTalk (the skill is the .agents/skills/voice-references/ folder at the pinned commit: SKILL.md only; the identical .claude/skills/ copy and the root skills/ copy are duplicates, not the source of this listing, and the rest of the NPCTalk repo is not installed) Pinned version: install the code at commit 59635ceb8ee010599c3caf9d9d1ee1aa114af91d exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/christos-antono/NPCTalk into a temporary folder, check out commit 59635ceb8ee010599c3caf9d9d1ee1aa114af91d, and copy the .agents/skills/voice-references/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 50e136554f7e458c1c7b6079970ca81f71dc5ad4ac4ef95282ebd19ab1f8cfcc If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds one file only: SKILL.md (4,346 bytes). There is no code, no script, no reference file and no dependency list in the folder: the skill is instructions for turning a rights-cleared recording into clean reference clips, a clone clip and mood clips, then checking and uploading them. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. The file names no telemetry. The use-time sends to fal and to your own Modal voice server, described next, happen only when the skill is used, never during this install. - Rights first, from SKILL.md: clone only voices you have the rights to: your own, an actor who agreed, a licensed or generated voice. A game's characters are voiced by real people; for a private single-player experiment you may decide to use the game's own audio, but it still never goes into a repo, a release or a public video without the rights, and npctalk check fails on audio files and .lab transcripts. - At use time, and only then, making the clips sends data to fal: SKILL.md says each candidate is transcribed with fal speech-to-text (FAL_KEY). In the npctalk CLI code that SKILL.md's commands run (npctalk/voice.py, outside this skill folder and not installed), each candidate WAV is sent as a POST to https://fal.run/nvidia/nemotron-asr-multilingual/asr with header "Authorization: Key $FAL_KEY" and the audio in the body as a base64 data URL (data:audio/wav;base64,...). That is a disclosure, not a step to run during this install: the audio excerpts from your recording leave your machine for fal at that point, on recordings you run. - Also at use time, uploading sends the finished voice folder to your own voice server: npctalk voice upload copies voices/<Name> (clone.wav / ref_*.wav clips and their .lab transcripts) into the server's Modal volume npcvoice-models at /voices/<Name> using the modal CLI (npctalk/voice.py, not installed). The server reads /voices in its Modal volume when it starts, so SKILL.md says to restart it after an upload (modal app stop npcvoice-drama; the next request starts a fresh one). Until a restart, a game can send the clip by URL with its transcript (reference_audio_url + reference_text), at about 9 s extra the first time. SKILL.md's check step also names a test line, npctalk say "Testing, one two." --voice Narrator/clone --out t.wav, which in npctalk/server.py (not installed) POSTs the line text and voice name to your own voice server (URL from NPCVOICE_URL, token NPCVOICE_TOKEN as an environment variable); do not run any of that during this install. - What voice make does, step by step in SKILL.md (the same steps built the GTA V example's 13 voices): split into lines at pauses of 0.3 s (20 ms frames louder than -42 dBFS are speech); join runs of up to 4 neighbouring lines into 6-11 s candidates and score them for normal level (about -20 dBFS, -15 for angry), even loudness, no clipping, few joins, about 9 s, keeping the best 12 that do not overlap; transcribe each with fal speech-to-text (FAL_KEY, disclosed above); drop candidates with fewer than 6 words or a speaking rate outside 1.4-4.8 words/s (screams, grunts, music, another language); write ref_0..2 (the best three) and clone.wav (the best candidates joined up to about 25 s), every clip with a .lab transcript next to it, because the cloner needs the exact words. Outputs: voices/<Name>/clone.wav + clone.lab (what the server clones from by default, as "Narrator/clone"), voices/<Name>/ref_0..2.wav + .lab (single clips for engines that want a short one), voices/<Name>/angry/ref_0..1.wav + .lab (used when a line starts with [angry]; keep 2 clips per mood), and voices_work/<Name>/cand_*.wav + .lab (every candidate, to listen to, never commit). Good source audio: one speaker, no music or effects under the voice, normal speaking level for the neutral voice (shouting goes in an angry mood folder, slurring in drunk), a minute or more of lines with pauses between them; if music is unavoidable, isolate vocals first (Demucs, as server/finetune/prepare_dataset.py does, outside this skill and not installed). - Which clip to clone from (measured in SKILL.md): on Drama 3, the about 30 s clone.wav gave 0.680 voice match on Lamar's test lines, each of his single 6-10 s clips gave 0.610-0.630, and passing all four clips as references at once was no better than clone.wav alone; S2 Pro preferred a single clean clip (0.707 vs 0.680), so test both if you switch models (skill voice-likeness-eval, not installed). Checks in SKILL.md: listen to clone.wav (anything in it, a laugh, a car, a second voice, shows up in every line), read the .lab (a wrong transcript makes the cloner mumble or skip words; fix the text by hand), say a test line (disclosed above), and for moods, if an angry line sounds like a different person, the mood clips are too far from the neutral ones (another recording session or mic): drop them. 3. When I say "yes, install", install it as a skill: copy the whole voice-references folder into the agent's skills directory, in a folder named "voice-references", keeping its file (SKILL.md). Do not use the repo's plugin, marketplace or npx routes: those install the whole NPCTalk plugin or all of its skills, not this single skill alone. 4. Verify with no network calls: frontmatter valid, file in place. 5. Report what was installed, where, and what I still need to do myself (for example: get a fal API key myself and set FAL_KEY as an environment variable, never in a file, only if and when I want to make clips; provide the npctalk CLI and ffmpeg myself, outside this skill; provide a recording I have the rights to myself; provide my own Modal voice server and its volume myself if I want to upload and use the clips, and restart it myself after an upload; using the skill is a separate step that only starts when I ask for it, and the use-time sends to fal and to my own Modal voice server disclosed above happen only then, on recordings I run). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.