Transcription serviceChinese
Chinese Transcription Services
Mandarin transcription with the character set, segmentation and number conventions fixed before the first file, staffed by annotators from the region the audio came from.
Free sample: Send a short Chinese audio clip and we return a transcript in the format and orthography you specify, at no cost, before you commit to a project.
Specification
- Audio
- Wideband, narrowband, mono, or separated channels
- Script
- Simplified or traditional characters chosen before the first file, punctuation width included
- Accuracy
- Target agreed per project and measured on a held-out slice of your own recordings
- Output
- Verbatim or clean-read text, timestamps, speaker turns
- Review
- Second-pass review on a sampled percentage, rate agreed up front
Dialects and varieties covered
Beijing Mandarin, Sichuanese.
Who does the work
- Contributor floor
- 1,000+per market
- Markets covered
- China
- Dialects covered
- 2dialects
Coverage is drawn from China, where Spirelight already publishes a delivered Mandarin Chinese dataset, so it is demonstrated rather than asserted.
Simplified or traditional is a delivery decision, and Cantonese is a different project
Written Chinese gives you two character sets for the same speech. Mainland audio is usually delivered simplified and Taiwan audio traditional, but that pairing is a convention rather than a rule, and converting afterwards is lossy wherever one simplified character stands for several traditional ones. Spirelight asks which set the delivery uses before the first file, and the answer goes in the style guide.
Cantonese is not Mandarin with an accent. Its vocabulary differs, its written form differs, and it is staffed and quoted as its own scope, which is why it has a Cantonese transcription page of its own in this cluster. The same holds for Shanghainese and Hokkien material. Naming it at scoping costs nothing, and finding it in QA costs a redelivery.
Where the words end is a judgment, because the writing does not say
Chinese text carries no spaces, so a transcript still has to settle how numbers, measure words, foreign names and full-width punctuation are written before anyone can compare two batches. Beijing speech adds the erhua suffix, which may or may not belong in the text depending on what you are training. Sichuanese audio brings vocabulary a Beijing annotator will guess at rather than know.
Those choices are set in the scope and applied by annotators assigned to the region in the recording. The accuracy target is measured on a held-out slice of your own audio, with an agreed rule for what counts as an error when a homophone is not resolved by the surrounding context.
This service is not the right fit when
- Cantonese, Shanghainese or Hokkien audio filed as Mandarin. Cantonese has its own page and its own pool here, and the other two are recruited on request, so naming the variety at scoping sends the files to speakers who understand them.
- Pinyin-only delivery. Transcripts are in characters unless a romanized layer is specified as an additional field.
- Machine-only transcription with no human pass, and translation into English.
Related
Need Chinese transcription scoped to your project? Send the audio conditions, volume, and turnaround you need. We will reply with a scoped quote and, if it helps, a free sample.