| %YAML 1.1 |
| %TAG !u! tag:unity3d.com,2011: |
| --- !u!114 &11400000 |
| MonoBehaviour: |
| m_ObjectHideFlags: 0 |
| m_CorrespondingSourceObject: {fileID: 0} |
| m_PrefabInstance: {fileID: 0} |
| m_PrefabAsset: {fileID: 0} |
| m_GameObject: {fileID: 0} |
| m_Enabled: 1 |
| m_EditorHideFlags: 0 |
| m_Script: {fileID: 11500000, guid: 4510294d23d964fe59443526f1ca7c4b, type: 3} |
| m_Name: Lip Sync |
| m_EditorClassIdentifier: |
| m_displayName: Lip Sync |
| m_hierarchyName: Lip Sync |
| m_context: {fileID: 11400000, guid: ccf5a8acd0168a6498a97e53cfd28d6a, type: 2} |
| m_markdownFile: {fileID: 0} |
| m_priority: 1012 |
| m_overrideMarkdownText: "# North Star\u2019s Implementation of ULipSync\n\nTo support |
| fully voiced NPC dialogue, NorthStar needed an efficient lip-syncing solution |
| that minimized reliance on animators. The team chose [ULipSync](https://github.com/hecomi/uLipSync) |
| for its familiarity, strong narrative control, customization, and ease of use.\n\n## |
| ULipSync Setup\n\nULipSync processes lip-sync data in three ways:\n\n1. **Runtime |
| processing** \u2013 Analyzes audio dynamically.\n2. **Baking into scriptable |
| objects** \u2013 Stores data for reuse.\n3. **Baking into animation clips** \u2013 |
| Prepares animations for timeline use.\n\nDue to CPU constraints and narrative |
| timelines, baking data into animation clips was the best approach.\n\n\n\n## |
| Phoneme Sampling & Viseme Groups\n\nULipSync maps phonemes (smallest speech components) |
| to viseme groups (facial animation controls).\n\n- English has **44 phonemes**, |
| but not all are needed for lip-syncing.\n- **Plosive sounds** (e.g., \"P\" or |
| \"K\") are hard to calibrate and may not impact the final animation significantly.\n- |
| Stylized models need fewer viseme groups than realistic ones, sometimes only |
| vowels.\n\nTo ensure flexibility, we recorded all 44 phonemes for each voice |
| actor, allowing system refinement later.\n\n\n\n## |
| Challenges in Phoneme Sampling\n\nNot all phonemes were sampled perfectly. Issues |
| included:\n\n- Regression effects, where certain phonemes worsened results.\n- |
| Lack of matching viseme groups, making some phonemes irrelevant.\n- Volume inconsistencies, |
| causing some sounds to be too quiet for accurate sampling.\n\nTo improve accuracy, |
| we documented problematic phonemes for future improvements and considered additional |
| recordings.\n\n## Ensuring Realistic Lip-Sync\n\nAutomated lip-syncing often |
| results in excessive mouth openness. Realistic speech involves frequent mouth |
| closures. To address this:\n\n- We referenced real-life speech patterns.\n- Animators |
| provided feedback to refine mouth movement accuracy.\n\n## Final Implementation\n\nEach |
| voice line was baked with a pre-calibrated sample array, storing blend shape |
| weights per NPC. This per-character approach worked due to a limited NPC count, |
| but a more generalized system is needed for larger projects.\n\n\n\n### |
| Relevant Files\n\n- [NpcController.cs](../Assets/NorthStar/Scripts/NPC/NpcController.cs)\n- |
| [NpcRigController.cs](../Assets/NorthStar/Scripts/NPC/NpcRigController.cs)\n" |
| m_overrideMarkdownRoot: .\Documentation/ |
|
|
