Chinese Transcription: Video & Audio to Text

How Chinese speech transcribes

Mandarin speech is recognised well. The transcript may occasionally switch between simplified and traditional characters, and homophones (many words sound alike) can be swapped, so names and technical terms deserve a check.

What a Chinese transcript looks like

Chinese transcripts are written in characters without spaces between words, using Chinese punctuation. Many words sound alike, so names and specialised terms are the main thing to check; the characters chosen for a name may not be the ones the person uses.

Examples

Costs: 40 credits (4¢) and 15 credits (1.5¢), including a translated transcript if you select another language. Audio files go in the audio to text workspace.

Simplified, Traditional, or a mix of both

Mainland China and Singapore read Simplified characters; Taiwan, Hong Kong and Macau read Traditional. A transcript can drift between the two, sometimes within one paragraph, so decide the script by audience and run the whole transcript through a script converter (OpenCC is a widely used open-source one) rather than fixing characters by eye. Conversion isn't purely mechanical: Taiwan and the mainland also differ in vocabulary, such as 软件 and 軟體 for software, or 视频 and 影片 for video, so have a reader from the target region check the result. Translating video for regional audiences discusses these regional choices.

Erhua, 二 and 两, and other spoken details

Beijing and other northern speakers add an r-sound to many words (这儿, 哪儿, 一点儿); southern and Taiwanese speakers rarely do. Check whether the transcript writes 儿 where the speaker said it, and whether you want it in published text, since formal writing often leaves it out. Numbers deserve a pass too. 二 and 两 both mean two but aren't interchangeable (两个人 but 第二), large figures may appear as 3万, 30000 or 三万, and percentages as 百分之五十 or in digits. Choose one style for the whole file.

English terms inside Mandarin sentences

Tech, finance and academic speakers drop English into Mandarin all the time: 这个 feature 下周上线, 我们要 follow up 一下. Check how these come out: in Latin letters, as a Chinese approximation, or missing. For published subtitles, decide whether English terms stay in Latin letters and whether you put a half-width space on each side; Chinese text has no spaces, so mixed lines can look cramped without them. Mandarin from Singapore and Malaysia may also include Hokkien, Cantonese or Malay words, which deserve the same check.

Punctuation conventions for Chinese subtitles

Running text uses full-width punctuation (，。？！：“”), and the transcript follows that. Subtitles are a different genre: many Chinese subtitle styles drop the full stop at the end of a cue and replace commas inside a cue with a space, keeping only question marks, exclamation marks and quotation marks. Check what your platform or client expects, then convert in one pass with find-and-replace. Watch for half-width commas or question marks mixed into Chinese text, which look slightly off on screen, and keep half-width punctuation for English phrases inside a line.

Searching and counting Chinese text

Searching a Chinese transcript for an exact character string works well, since there are no spaces to get in the way. The difficulty is homophones: if a name was written with the wrong characters, searching for the right ones finds nothing. When a term seems missing, search for each character separately or for a likely homophone. For word counts, frequency lists or research coding, run the text through a segmentation tool such as jieba first, because a raw character count tells you little about vocabulary.

Pinyin and tones aren't part of the output

The transcript is characters only. If you're preparing material for learners, add pinyin with a separate tool and check the readings of characters with more than one pronunciation, such as 行 (xíng or háng) and 长 (cháng or zhǎng), which automatic tools can get wrong in context. Learning a language with videos shows how learners use a transcript alongside the video. For translated text and English bylines, mainland names follow Hanyu Pinyin, but Taiwanese and Hong Kong names often use other romanizations, so take spellings from the person's own usage.

A 65-minute livestream replay, costed

File

65-minute MP4 replay of a Mandarin product launch

Transcript

65 credits (6.5¢), Chinese selected (no translation)

Second run

English as the target, another 65 credits (6.5¢)

Script decision

Converted to Traditional for a Taiwan distributor, then read by a Taiwanese colleague

Subtitle edit

Long cues split at the comma between clauses, English product names kept in Latin letters

The YouTube subtitles page covers uploading the finished SRT.

Using the tool on this page

• The spoken language is detected automatically; you select the output language — the spoken language itself for an untranslated transcript, or another language for a translation. Files can be up to 2 hours long.
• Only the audio is sent for processing — the video stays on your device — and audio and text are deleted within 30 minutes of delivery (how files are protected).
• You can download a timestamped transcript, subtitles (SRT and VTT) and plain text. Every output, step and limit is explained on the video to text page.