引導式測驗 · 中等 等級
人工智慧在唇讀和視覺語音辨識測驗的應用
測試唇讀模型中的視覺語音提示、視位歧義、音訊視訊融合、時間特徵、嘴部區域處理和 CTC 對齊。
問題 1 的 8
視位分組在視覺語音辨識中描述什麼歧義?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- 視位分組在視覺語音辨識中描述什麼歧義?
- Every sound has a uniquely recoverable image in every frame
- A camera directly measures a word's meaning
- Different speech sounds can share similar visible mouth movements
- The speaker's filename determines the phoneme sequence
- 當噪音增加時,視覺流如何幫助視聽辨識器?
- It guarantees correct recognition regardless of occlusion or lighting
- It can supply additional visible articulation cues when the audio is degraded
- It removes every need to evaluate the fusion model
- It makes the acoustic signal physically noiseless
- 唇讀模型中的 3D 卷積前端有助於捕捉什麼?
- 說話者眼睛的顏色
- 室溫
- 說話者寫作時的口音
- 連續幀之間的短期運動模式
- 哪些擷取條件可以隱藏視覺語音模型所需的資訊?
- Head turns, poor lighting, or an obstruction over the mouth
- A clear frontal view as the only guaranteed failure condition
- A rule that higher resolution removes all ambiguity
- An audio file's compression setting with no effect on the video
- 原始LipNet句子辨識結果該附加什麼評估範圍?
- Every spontaneous real-world conversation in every language
- The constrained GRID benchmark used in the study
- A guarantee that all professional lip readers are always outperformed
- Only written text with no visual speech input
- 社會對唇讀人工智慧的主要擔憂是:
- 它用電量太多
- 它無法在計算機上運行
- 秘密使用可以在未經同意的情況下進行監視
- 它僅適用於一種語言
- Why might a visual-speech pipeline crop a mouth-centered region before temporal modeling?
- To focus processing on relevant articulation rather than unrelated image background
- To remove the need for any information about movement
- To guarantee that every phoneme becomes visually unique
- To supply the final transcript directly from the crop coordinates
- What alignment requirement does CTC avoid in the LipNet training setup?
- All transcript-level training targets
- All observations of the speaker's mouth
- Every temporal dependency in the recognition problem
- A supplied character label for every individual video frame