Skip to content

docs(genai,#18062): tranche Audio/04 — VRAM 04-6/04-13, services 04-7, presentation 04-16 #18071

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
myia-ai-01 merged 1 commit into from
Sep 28, 2026
Merged

docs(genai,#18062): tranche Audio/04 — VRAM 04-6/04-13, services 04-7, presentation 04-16 #18071

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 7 additions & 4 deletions MyIA.AI.Notebooks/GenAI/Audio/04-Applications/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,26 +36,29 @@ Pipeline complet de 7 notebooks pour générer un audiobook à partir d'un texte

| # | Notebook | Pass | Contenu | Service | VRAM |
| --- | ---------- | ------ | --------- | --------- | ------ |
| 6 | [04-6-Audiobook-Pipeline](04-6-Audiobook-Pipeline.ipynb) | P0 | Pipeline audiobook orchestrateur | Mixed | ~10 GB |
| 7 | [04-7-TTS-Voice-Benchmark](04-7-TTS-Voice-Benchmark.ipynb) | P0 | Benchmark comparatif des modèles TTS | Kokoro + OpenAI | ~2 GB |
| 6 | [04-6-Audiobook-Pipeline](04-6-Audiobook-Pipeline.ipynb) | P0 | Pipeline audiobook orchestrateur | Mixed | ~5 GB |
| 7 | [04-7-TTS-Voice-Benchmark](04-7-TTS-Voice-Benchmark.ipynb) | P0 | Benchmark comparatif des modèles TTS | Kokoro + Qwen3 TTS + OpenAI | ~2 GB |
| 8 | [04-8-Lecture-Analytique](04-8-Lecture-Analytique.ipynb) | P1 | Analyse littéraire, segmentation dialogues/narration | OpenAI API | 0 |
| 9 | [04-9-Voice-Casting](04-9-Voice-Casting.ipynb) | P2 | Attribution de voix par personnage, casting vocal | OpenAI API | 0 |
| 10 | [04-10-Annotation-Prosodique](04-10-Annotation-Prosodique.ipynb) | P3 | Tags prosodiques FishAudio S2-Pro | OpenAI API | 0 |
| 11 | [04-11-Generation-TTS](04-11-Generation-TTS.ipynb) | P4 | Génération audio Kokoro multi-voix | Kokoro TTS | ~2 GB |
| 12 | [04-12-Compilation-Audio](04-12-Compilation-Audio.ipynb) | P5 | Concaténation FFmpeg + normalisation loudness | FFmpeg | 0 |
| 13 | [04-13-Audiobook-FishAudio-S2Pro](04-13-Audiobook-FishAudio-S2Pro.ipynb) | Full | Pipeline v4 FishAudio S2-Pro, 29 tags prosodiques, validation WER | FishAudio + Whisper | ~2 GB |
| 13 | [04-13-Audiobook-FishAudio-S2Pro](04-13-Audiobook-FishAudio-S2Pro.ipynb) | Full | Pipeline v4 FishAudio S2-Pro, 29 tags prosodiques, validation WER | FishAudio + Whisper | ~6 GB |

### Hommage Munkres — musique et mathématiques

| # | Notebook | Contenu | Service | VRAM |
|---|----------|---------|---------|------|
| 14 | [04-14-VoiceLeading-Rendu-GenAI](04-14-VoiceLeading-Rendu-GenAI.ipynb) | Rendu audio intelligent du voice leading App-21 : baseline téléphone vs MusicGen-melody (3 styles H1), spectrogrammes mesurés | MusicGen-melody (GPU local) | ~6 GB |

### Famille YuE2 — compréhension musicale
### Famille YuE2 — compréhension et transcription musicales

Trois satellites : la **compréhension** (MERT2, 04-15), la **génération** ([YuE2, 02-7](../02-Advanced/02-7-YuE2-Song-Generation.ipynb)) et la **transcription** (SheetSage2, 04-16).

| # | Notebook | Contenu | Service | VRAM |
|---|----------|---------|---------|------|
| 15 | [04-15-MERT2-Music-Understanding](04-15-MERT2-Music-Understanding.ipynb) | Embeddings MERT2 (632 M, self-supervised, poids CC BY-NC 4.0) : corpus contrôlé 3 motifs × 4 timbres × 2 transpositions, similarité cosinus, retrieval contexte long FullSong (medley 120 s), sondes linéaires LOO motif vs timbre sur couches L1/L23/L24 | Local GPU (kernel `mert2-gpu`, pins torch 2.6.0 / transformers 4.53.2) | ~4 GB |
| 16 | [04-16-SheetSage2-Audio-To-Score](04-16-SheetSage2-Audio-To-Score.ipynb) | Transcription audio vers partition éditable : SheetSage2 (677 M, base MERT-v2-FullSong + adaptateurs), cinq têtes (mélodie, accords, temps forts, tonalité, structure), artefacts ABC/MIDI/événements horodatés, comparaison à vérité-terrain connue par construction, cycle symbolique→audio→symbolique mesuré | Local GPU (kernel `sheetsage2-gpu`) ; poids CC BY-NC 4.0 (usage commercial interdit) | ~8 GB |

> **Note (consolidation #13741)** : un sibling pédagogique plus minimal (re-voicing CP-SAT + MIDI synthèse, 9 cellules, 51 Ko) a été archivé sous [`MyIA.AI.Notebooks/GenAI/Audio/_archive/04-14-VoiceLeading-RenduGenAI.ipynb`](../_archive/04-14-VoiceLeading-RenduGenAI.ipynb) pour éliminer le doublon de slot 04-14. Le contenu reste préservé (re-voicing, synthèse MIDI, MusicGen, spectrogrammes, export WAV) ; ce notebook canonique `Rendu-GenAI` (5.16 Mo, 27 cellules, version complète re-exécutée #13398) reste la référence.

Expand Down
Loading