英文语音识别标注标准

English speech recognition labeling standard

语音识别（ASR）指把语音转换成文字。任务是把音频中的speech（说话）一字不落的标注出来。

Speech recognition (ASR) refers to the conversion of speech into text. The task is to mark out the speech in the audio without dropping a word.

1. 登录小核众测官网https://zc.bytedance.com/，点击更多任务；

Log on to the small nuclear public site https://zc.bytedance.com/, click more tasks;

2. 搜索ASR并点击该队列，点击开始任务；

Search for ASR and click the queue, click start task;

3. 标注流程：

Annotation process:

4. 语音类型判断标准：

Speech type criteria:

1) speech：可听清的人说话声，若视频中有多人说话，需要都写出来；若音频中有部分时段多人说话声重叠，且很清晰，需要把重叠部分截掉（rap：节奏感不是很强的，也可以标注。）

Speech: can hear the sound of people talking, if there are many people in the video speak, need to write it; If there are parts of the audio with multiple voices overlapping and clear, the overlapping parts need to be cut off (rap: rhythm is not very strong, can also be tagged.)

2) 非speech：音乐、唱歌 、动物叫声和自然界的声音

Non-speech: music, singing, animal calls and natural sounds

3) 丢弃：英语除外的其他语种、听不清、嘈杂声

Discarded: other languages other than English, inaudible, noisy

5. 文本书写标准：

Text writing standards:

1) 不加标点符号，单词间需加空格

Without punctuation, spaces should be added between words

2) 专有名词、人名、电影名、书名 每个单词首字母大写；缩略语每个字母都需大写，其余都小写（包括句子首字母第一个单词）

Proper nouns, personal names, movie names, book titles, each word is capitalized; acronyms are capitalized for each letter, and the rest are lowercase (including the first word in a sentence)

3) 数字不要写阿拉伯数字，比如，59--fifty-nine

Numbers don't write Arabic numerals, for example, 59--fifty-nine

4) 若单词发一半，可以不写

If the word is half pronounced, you can leave it

5) 正常按照音频发音标注，若用户发音错误，需要按正确的标注出来

Note normally according to the audio pronunciation, if the user pronunciation is wrong, it is necessary to mark it correctly

6) 邮箱和网址按照正常形式输出，比如：www.yahoo.com

Mailboxes and URLs are exported in normal form, such as: www.yahoo.com

6. 截取操作

Interception operation

1) 需要截取的情况：句首或句尾有听不清的语音、嘈杂音、静音、多人说话重叠等需截掉

Situations where interception is required: inaudible sounds, noise, mute, overlapping of speech, etc., at the beginning or end of a sentence

2) 截取方式：可通过点击【截取开始】和【截取结束】选定截取区间（或者对应的快捷键），然后点击【截取确认】（或者使用快捷键a或5），此时区间内的语音将自动播放，表示截取完成

Interception: select the intercept interval (or the corresponding shortcut key) by clicking on [intercept start] and [intercept end], and then click [intercept confirmation] (or use shortcut key a or 5), where the voice within the interval will be played automatically, Indicates completion of interception

3) 截取技巧：拖动小红点进行截取区间修改，点击上方波形图可显示小红点

Interception technique: drag small red dot to modify the intercept interval, click on the above waveform to display the small red spot

4) 注意： 截取后要确认一下语音和文本是否对应

Note: after intercepting, verify that the speech and text correspond

5) 必须在原截取区间内截取，比如原语音的播放区间为3-8s，只能在3-8s内截取，不可截长至1-8s

Must be intercepted within the original intercept interval, for example, the playback interval of the original speech is 3-8s, can only be intercepted within 3-8s and cannot be cut to 1-8s

7. 快捷键

1) 空格-提交Spaces-submission

2) 1-开始1-start

3) 2-暂停2-suspension

4) 5-重复播放截取区间5- repeat play intercept interval

5) q-丢弃Q-discard

6) w-非speechW-non-speech

7) s-截取开始S- start of interception

8) e-截取结束E- end of interception

9) a-截取确认A-intercept confirmation

10) shift+alt-文本切换Shift alt- text switching

7. 部分技巧

Partial technique

1) 多使用快捷键Use shortcuts more often

2) 可以先理解视频大概意思再标注

You can understand the general meaning of the video and then annotate it.

3) 可以根据意群，标注

Can be tagged according to the meaning group

4) 对一些出现率高的视频语音进行文本整理，可直接粘贴复用

Text finishing for some video voice with high frequency, which can be directly pasted and multiplexed