Can You Transcribe Mixed Chinese–English Audio on Mac?

The short answer

You can attempt a transcript of a recording that contains both Chinese and English, but automatic language detection is not a guarantee that every switch will be preserved correctly. Start with a representative excerpt, keep translation off while checking the original words, and review the places where the speaker changes language.

Alison Workspace provides local file transcription and editable text on Apple-silicon Macs running macOS 15 or later. Its available recognition paths and resources affect the result. This guide explains how to evaluate the output and recover difficult passages; it does not claim a measured accuracy rate or reliable recognition of every mixed-language sentence.

Identify which multilingual problem you have

Three jobs often get described as “bilingual transcription”:

• Several files, each mainly in one language — What needs to happen: Select or detect the language for each file or compatible batch
• One recording that switches between Chinese and English — What needs to happen: Preserve the spoken words across language changes
• A Chinese recording that needs an English reading copy — What needs to happen: Transcribe the original, then translate it

This guide focuses on the second job. Producing a Chinese transcript with an English translation is useful, but it does not prove that an original English phrase in the audio was transcribed faithfully.

The pattern of switching matters too. A speaker may change language after a paragraph, borrow an English product name, or alternate languages inside one sentence. A setup that handles long separate passages may still miss short words embedded in another language.

Choose an excerpt that exposes the problem

Use a recording you are allowed to process. Preserve the original and prepare a short review copy with your usual audio editor if necessary. Include a clear example of each language and at least one actual transition; an opening monologue in only one language is not a representative test.

Write a small reference for the difficult passage by listening carefully. Mark uncertain words rather than inventing them. Check names against reliable accompanying material when possible. The reference does not need to cover the entire recording to reveal whether an English phrase disappears or a Chinese sentence becomes an unintended translation.

A review copy has its own timeline. If you cut a passage from the middle of a recording, note its starting position in the original. Its exported timestamps will refer to the excerpt, not automatically to the full recording.

Prepare recognition without adding translation

In Alison, open Transcribe Files…, add the supported recording, and check its displayed duration. Prepare the recognition resources requested for the path you intend to use. File transcription reads the recording directly; it does not require speaker playback or microphone input.

Leave translation off for the initial check. This makes it easier to see whether the source transcription preserved what was spoken. A translation target is a separate output choice, not a list of all the languages present in the audio.

If the app offers automatic recognition with its required model ready, try that on the excerpt. Where a recording is mostly one language with occasional borrowed terms, a separate trial with the known main language can also be informative. Compare the actual output; neither choice should be presented as universally better for mixed speech.

Record the app version, macOS version, recognition choice, and any optional engine mode. Changing several settings at once makes it difficult to tell which change affected the result.

Do not equate automatic detection with perfect switching

Automatic detection selects a language from the audio available to the recognizer. Different processing paths can examine different amounts of audio and make different routing decisions. The live-listening path and file-transcription path should not be assumed to behave identically.

Alison's public listing describes automatic language changes during listening. That is not a benchmark for every mixed-language file. Likewise, a model being described as multilingual means it can work across languages; it does not establish reliable recognition of every within-sentence switch.

In the upstream Whisper project, a maintainer's explanation identifies code-switching as a limitation and discusses separating passages for processing. That is useful background, not evidence that Alison uses precisely the same routing or has been tested on your recording.

Judge the result at the transitions. A convincing first paragraph is not enough.

Inspect words, language, and completion separately

Compare the excerpt with the source audio and your reference:

• Did both Chinese and English survive in the source text?
• Were any passages translated instead of transcribed as spoken?
• Were short English terms replaced with similar-sounding words?
• Are names, numbers, negative statements, and conditions correct?
• Did the transcript reach the end, or is it partial?

For example, “这个版本先不要 deploy” includes a Chinese negative instruction and an English verb. A transcript that loses “不要” changes the instruction; one that substitutes an unrelated English word creates a different error. This is an illustrative review example, not a reported test result.

Also distinguish script conventions from lost speech. Differences in punctuation or Chinese script presentation do not always mean words were omitted. Check the audio and intended output before “correcting” a faithful term into something else.

Recover difficult passages deliberately

If long language changes cause trouble, use an external audio editor to make copies of meaningful passages, then transcribe each with suitable settings. This is a manual fallback, not a promise that Alison automatically splits a recording by language or merges the results back into one timeline.

Keep enough surrounding speech to preserve context. Cutting every borrowed English word into a tiny clip can remove the information needed to understand it. Sentence-internal switching may still need careful manual correction even after you change settings.

For a document, you can combine reviewed passages in a text editor and retain notes about where each came from. For SRT or VTT, combining files is more demanding: each excerpt's times need to be aligned to the original media. Do not simply concatenate subtitle files and assume synchronization remains correct.

If different speakers use different languages, speaker labels may help organize review. They do not make a voice's language permanent, and diarization alone does not solve switching by the same speaker.

Translate only after the original is useful

Once the source text preserves the words you need, decide whether a translated copy would help. Check the language pairs actually available on the Mac and review mixed passages individually. A single paragraph containing two languages may need more care than a paragraph entirely in one source language.

Keep the reviewed source alongside any translation. If a translation was created before source corrections, translate the reviewed version again and inspect the changed lines. Check coverage as well as fluency: a partly translated document can look complete at a glance.

The bilingual transcript guide covers source review, translation coverage, and export in more detail.

Export a reviewed record and retain its context

Choose original text when the task is to preserve Chinese and English exactly as spoken. Choose bilingual output when you also need an available translation; check that the output does not hide missing translated passages.

TXT or Markdown is useful for further editing, while PDF provides a reading copy. Open the saved file and inspect Chinese characters, English names, the final passage, and any speaker labels or timestamps you included. Keep the original recording available for later verification.

After preparation, Alison's supported processing runs locally, but an exported file saved in a synced or shared folder can still leave the device through that service. Choose a destination appropriate for the recording and the people involved.

The first milestone is a short passage you have checked carefully. Expand to the full recording only when you understand which parts work well and which still require review.

References checked October 28, 2026: the linked Whisper maintainer discussion and Alison Workspace's public description. Product-path distinctions were also reviewed in project documentation and source. No mixed-language recording benchmark was performed for this article; the examples describe checks for the reader to carry out.