Exécutez n'importe quel Skill dans Manus
en un clic

Exécutez n'importe quel Skill dans Manus en un clic

audiocpp-diarize

Étoiles46

Forks6

Mis à jour20 juillet 2026 à 18:58

Speaker diarization: figure out who spoke when in a recording and produce a speaker-labeled transcript, by spawning a private audio.cpp engine with the sortformer diarization model and labeling each turn with the local ASR engine. Use when the user asks to diarize a meeting or call recording, separate or identify speakers, or wants a transcript broken down by speaker.

Installation

Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.

Exécuter dans Manus

Source

Elumenotion

Elumenotion/GuideAnts

Ouvrir le dépôt GitHub Voir les dépôts du créateur

Téléchargement

Exécuter dans Manus

Métiers associésSOC

Basé sur la classification professionnelle SOC

Développeurs de logicielsProfessions informatiques et mathématiques·SOC 15-1252

Explorateur de fichiers

5 fichiers

SKILL.md

readonly

name	audiocpp-diarize
description	Speaker diarization: figure out who spoke when in a recording and produce a speaker-labeled transcript, by spawning a private audio.cpp engine with the sortformer diarization model and labeling each turn with the local ASR engine. Use when the user asks to diarize a meeting or call recording, separate or identify speakers, or wants a transcript broken down by speaker.
metadata	{"guideants":{"enabled":true,"display_order":34,"requires_toolsets":["sandbox"]}}

audio.cpp speaker diarization (experimental)

GuideAnts has no local diarization (only its Azure cloud transcription provider does it), but the container's audiocpp_server binary ships the sortformer_diar loader and a generic /v1/tasks/run route. This skill downloads the model once, spawns a private engine, and turns any recording into a speaker-labeled transcript. Deliverables land in Output/; the live voice path is untouched. E2E-verified in this stack (2026-07-15).

Preflight

python3 Output/Skills/audiocpp-diarize/scripts/preflight.py --for diarize

Trust its verdict over this document. It also tells you whether the model is already on disk (skip the download) and whether ASR labeling is available.

The recipe

# 1. Download the model — skip if preflight says modelAlreadyDownloaded
python3 Output/Skills/audiocpp-diarize/scripts/fetch_model.py nvidia/diar_sortformer_4spk-v1 \
  --dest /models-local/asr/diar_sortformer_4spk-v1 --exclude diar_sortformer_4spk-v1.nemo

# 2. Spawn the private engine (loads in seconds; ~494 MB weights, fits next to
#    the wrappers' models without an unload)
python3 Output/Skills/audiocpp-diarize/scripts/spawn_engine.py start \
  --path /models-local/asr/diar_sortformer_4spk-v1 --family sortformer_diar --task diar

# 3. Diarize (any ffmpeg-decodable input)
python3 Output/Skills/audiocpp-diarize/scripts/diarize.py Output/uploads/meeting.mp3 -o Output/meeting

# 4. Always stop the engine when done
python3 Output/Skills/audiocpp-diarize/scripts/spawn_engine.py stop

diarize.py converts the input to the 16 kHz mono WAV the engine requires (the raw engine does not resample), fetches speaker turns, merges/filters them, and — when the GuideAnts ASR engine is up — transcribes each turn. Outputs: <base>.transcript.txt (speaker-labeled, timestamped) and <base>.diarization.json (structured turns with confidences).

Options that matter

--turns-only — skip ASR labeling (or it degrades to this automatically, with a note, when no ASR model is loaded).
--language <hint> — passed to the ASR engine per turn.
--merge-gap-seconds / --min-turn-seconds — shape the output without touching the engine.
Engine-side tuning needs a re-spawn with e.g. --option speaker_threshold=0.4 (also: speaker_min_frames, speaker_pad_frames, session_len_sec).

Limits — say these to the user when relevant

Offline only; at most 4 speakers (model variant).
Speaker ids (SPEAKER_00…) are arbitrary labels — ask the user who is who if they want real names in the deliverable.
Long recordings: per-turn labeling can hit the ~5-minute script budget; the JSON then has "labeling": {"partial": true} — rerun diarize.py (the diarization pass itself is fast; only labeling resumes slowly).

Reporting

End by telling the user what worked and what was blocked, quoting the preflight/engine evidence.

Plus depuis ce dépôt

même dépôt

audiocpp-asr-extended

Elumenotion/GuideAnts

Transcription beyond the GuideAnts wrapper contract: transcribe workspace files by path (no upload, no 50 MB gateway cap), pass language hints, and sideload other qwen3-family ASR snapshots from Hugging Face through the wrapper. Use when a transcription needs a language hint, the file is large, or the user wants to try an ASR model the catalog doesn't list.

2026-07-2046

audiocpp-deferred-tts

Elumenotion/GuideAnts

Run TTS model families audio.cpp supports but GuideAnts does not ship — Qwen3 CustomVoice builtin speakers, VibeVoice multi-speaker dialogue, MioTTS, VoxCPM2, PocketTTS, Vevo2 voice conversion — by downloading the model and spawning a private audiocpp_server in the sandbox. Use when the user names one of these models or asks for a TTS capability the loaded catalog model rejects.

2026-07-2046

audiocpp-host-tts

Elumenotion/GuideAnts

Synthesize speech against the user's own host-native audiocpp_server build (outside the GuideAnts container) — including model families the container binary lacks, like Kokoro or Parakeet forks. TTS only. Use when the user says they run their own audio.cpp server or asks for a family the container build cannot load.

2026-07-2046

audiocpp-extended

Elumenotion/GuideAnts

Experimental: use audio.cpp scenarios and models that GuideAnts does not ship — speaker diarization (who spoke when, speaker-labeled transcripts), voice cloning from user-supplied audio, deterministic seeds, builtin speakers, language forcing, and deferred TTS families (Qwen3 CustomVoice, VibeVoice, MioTTS, VoxCPM2, PocketTTS) — by calling the raw audiocpp_server engines directly or spawning a private engine in the sandbox. Use when the user asks for a voice, model, or audio scenario the built-in audio tools reject or cannot express, including diarizing a meeting recording or labeling speakers in a transcript.

2026-07-2046

audiocpp-tts-controls

Elumenotion/GuideAnts

Advanced synthesis controls on the loaded audio.cpp TTS model: deterministic output via seed, forcing the spoken language, voice-design from a text description (instructions), and enumerating builtin speaker ids — none of which the built-in GuideAnts audio tools expose. Use when the user wants reproducible audio, a specific language, a described voice, or a list of available speakers.

2026-07-2046

audiocpp-voice-clone

Elumenotion/GuideAnts

Clone a voice from a user-supplied reference clip and synthesize speech with it, by calling the raw audio.cpp TTS engine with voice_ref/reference_text — fields the built-in GuideAnts audio tools do not expose. Use when the user wants generated speech to sound like their own voice or another consenting speaker's recording.

2026-07-2046

name	audiocpp-diarize
description	Speaker diarization: figure out who spoke when in a recording and produce a speaker-labeled transcript, by spawning a private audio.cpp engine with the sortformer diarization model and labeling each turn with the local ASR engine. Use when the user asks to diarize a meeting or call recording, separate or identify speakers, or wants a transcript broken down by speaker.
metadata	{"guideants":{"enabled":true,"display_order":34,"requires_toolsets":["sandbox"]}}

audio.cpp speaker diarization (experimental)

Preflight

python3 Output/Skills/audiocpp-diarize/scripts/preflight.py --for diarize

Trust its verdict over this document. It also tells you whether the model is already on disk (skip the download) and whether ASR labeling is available.

The recipe

# 1. Download the model — skip if preflight says modelAlreadyDownloaded
python3 Output/Skills/audiocpp-diarize/scripts/fetch_model.py nvidia/diar_sortformer_4spk-v1 \
  --dest /models-local/asr/diar_sortformer_4spk-v1 --exclude diar_sortformer_4spk-v1.nemo

# 2. Spawn the private engine (loads in seconds; ~494 MB weights, fits next to
#    the wrappers' models without an unload)
python3 Output/Skills/audiocpp-diarize/scripts/spawn_engine.py start \
  --path /models-local/asr/diar_sortformer_4spk-v1 --family sortformer_diar --task diar

# 3. Diarize (any ffmpeg-decodable input)
python3 Output/Skills/audiocpp-diarize/scripts/diarize.py Output/uploads/meeting.mp3 -o Output/meeting

# 4. Always stop the engine when done
python3 Output/Skills/audiocpp-diarize/scripts/spawn_engine.py stop

Options that matter

--turns-only — skip ASR labeling (or it degrades to this automatically, with a note, when no ASR model is loaded).
--language <hint> — passed to the ASR engine per turn.
--merge-gap-seconds / --min-turn-seconds — shape the output without touching the engine.
Engine-side tuning needs a re-spawn with e.g. --option speaker_threshold=0.4 (also: speaker_min_frames, speaker_pad_frames, session_len_sec).

Limits — say these to the user when relevant

Offline only; at most 4 speakers (model variant).
Speaker ids (SPEAKER_00…) are arbitrary labels — ask the user who is who if they want real names in the deliverable.
Long recordings: per-turn labeling can hit the ~5-minute script budget; the JSON then has "labeling": {"partial": true} — rerun diarize.py (the diarization pass itself is fast; only labeling resumes slowly).

Reporting

End by telling the user what worked and what was blocked, quoting the preflight/engine evidence.