引導式測驗 · 中等 等級
聲音事件偵測測驗
研究聲音類別和時間邊界,包括複調、弱標籤、DCASE 基線、起始容差和缺失類別註釋。
問題 1 的 8
除了片段層級的標籤外,時間聲音事件偵測還估計了哪些額外的輸出?
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。
- 除了片段層級的標籤外,時間聲音事件偵測還估計了哪些額外的輸出?
- Only the audio file’s compression ratio
- Only its filename’s alphabetical order
- Event onset and offset boundaries within the clip
- Only the display color of a waveform
- 為什麼聲音事件檢測經常被描述為複調問題?
- 因為它只處理音符
- 因為多個聲音可以重疊同時出現
- 因為它需要多個麥克風
- 因為音訊始終是立體聲
- SED 訓練資料中的「弱標籤」是什麼意思?
- 標籤已加密
- 音頻品質很差
- 僅標記安靜的聲音
- 剪輯被標記為事件發生,但不準確是在何時發生
- 在引用的 DCASE 2024 基準中指出了哪種分類器結構?
- A convolutional recurrent neural network with added pretrained frame features
- Only a static spreadsheet sort
- Only an unchanged relational database table
- Only a mandatory image-file compressor
- 此處描述的記錄聲音偵測基準情境中使用了哪種時頻表示法?
- Only a speech transcript
- A log-mel representation
- Only a JPEG photograph
- Only an unchanged output filename
- Which cited research challenge supplies the 2018 and 2024 sound-event task settings discussed here?
- 奧運
- A retail shopping holiday
- A mandatory social-network ranking
- DCASE
- Considering only the DCASE 2018 onset collar of 200 ms, does a 1.15 s prediction satisfy it for a 1.00 s reference onset?
- Yes; the onset error is 150 ms
- No; the error is 1.15 s
- No; every predicted onset must be numerically identical
- Yes; onset errors are never evaluated
- In the heterogeneous 2024 task, how does the baseline handle outputs for classes whose annotations are missing in a training dataset?
- Treats every unannotated class as certainly absent
- Copies all missing labels from the final test answers
- Masks those unavailable outputs in the relevant loss computation
- Eliminates every labeled class from the model