引導式測驗 · 中等 等級
Mozilla 通用語音資料集測驗
測試 Common Voice 腳本化和自發性剪輯、志願者驗證、發布/版本追蹤、揚聲器分割、CC0 範圍和隱私。
問題 1 的 8
為什麼要在實驗中錄製 Common Voice 發行版本?
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。
- 為什麼要在實驗中錄製 Common Voice 發行版本?
- 版本包含完全相同的剪輯
- 資料集版本之間的內容和覆蓋範圍可能會發生變化
- 版本控制證明每個註釋都是正確的
- 版本識別碼取代了拆分文檔
- 腳本剪輯的志工驗證會檢查什麼?
- 說話者的合法身份
- 保證的未來 ASR 性能
- 各方言的代表性
- 音訊是否與顯示的提示相符
- 為什麼來自一位貢獻者的多個剪輯應保留在一個訓練測試分區中?
- 共同的聲音特徵可能會滲透到評估中
- All clips have the same cryptographic digest
- Each script contains only one sentence
- One speaker can contribute only one recording
- What use claim does a good score on scripted Common Voice speech not establish?
- The model could transcribe prompted sentences
- A versioned dataset was used
- Equal quality on spontaneous noisy conversations
- Scripted clips can be scored with references
- Where does Mozilla make current Common Voice releases available?
- Only a private vendor application
- Mozilla Data Collective under release-specific terms
- Only synthetic speech files
- Only an unlabeled image archive
- Which privacy fact remains true for public voice data?
- A recording can still carry identifying speech characteristics
- Public audio is inherently anonymous
- CC0 prevents any voice resemblance
- All optional metadata are complete
- How should CC0 be interpreted for a particular Common Voice release?
- Every audio dataset worldwide is CC0
- Public licensing erases privacy concerns
- Consent rules no longer matter
- Check the release terms; Mozilla says CC0 unless otherwise specified
- What extra evaluation is needed before using a model in live calls?
- Only clean prompted-speech WER
- Only the clip count of one release
- Testing on representative conversational audio from the intended setting
- Only the distribution page description