To translate audio to text on Mac, first turn the recording into words in its spoken language. Correct those words against the audio, then translate the corrected text into the language you need. Keep the recording, source transcript and translation as three separate files so you can trace a questionable sentence back to what was said.
A transcript preserves the spoken language. A translation changes the language. A dubbed recording changes the spoken audio. If your deliverable is a readable document in another language, a source transcript followed by written-text translation gives you a clear place to check each operation. You do not need to create a new voice track.
The crucial review happens before translation. Fluent target-language prose can still carry an incorrect name, missing negative or wrong number from the source transcript. Fixing the wording afterward is harder when you no longer know which step introduced the mistake.
We develop and sell Paraspeech. Paraspeech is not affiliated with or endorsed by Google. Documentation checked October 13, 2026; the examples below are authored illustrations, not measured translation results.
Transcribe the recording in its spoken language
Start by identifying the language you hear and the language your reader needs. Write both at the top of your working document. A recording in Spanish for an English-speaking colleague needs a Spanish source transcript and an English translation; choosing English as the recognition language would confuse those jobs.
Choose a transcription route that supports the recording's language and file. Our audio-file transcription guide covers that first decision. Keep the original audio unchanged and work from a copy. If the recording contains several languages, check the selected model's documented coverage rather than assuming one language setting handles every passage.
Paraspeech's released direct-Mac file workflow can supply the source words on supported Apple Silicon with a compatible downloaded offline model and eligible access. Import the file, wait for processing, then use the per-item transcript copy control. File and language compatibility depend on the selected model; Intel cloud dictation is a different route, not this local file-import path.
The copied text is the text to review, not proof of a verbatim record. Listen back before using the words as the translation source. Paraspeech also offers per-item VTT export, which is useful when you need timed text; VTT and SRT serve subtitle workflows, while an ordinary working document is easier for paragraph translation.
Download Paraspeech if a local Mac file-transcription stage fits your recording. The offline processing guide explains the model setup. Local audio recognition and the later Google Translate browser step are separate processing choices; this whole workflow does not stay offline.
Correct the source words before translating
Listen through the source transcript with the recording available. Mark uncertain names, numbers and instructions. If you cannot hear a word confidently, leave a visible uncertainty note and return to that passage; a translator cannot recover a sound that the transcript omitted.
Here is a synthetic example, not output from either app. Imagine the spoken instruction is: “Do not renew if the price exceeds $80.” A source transcript that drops “not” changes the instruction to renew under that condition. Translating the incorrect sentence faithfully would preserve the wrong instruction in another language.
The source review therefore has a concrete purpose: settle what was said before deciding how to express it for another reader. Preserve meaningful hesitation or qualification when it changes the message. A speaker saying “possibly Friday” has not promised Friday, even if a cleaner sentence sounds better without the first word.
Save the corrected source as a separate version. Give the files names that distinguish their roles, such as interview-source-reviewed.txt and interview-target-draft.txt. Those are example filenames, not exported files from a performed test. They let you correct a later translation without losing the words you checked against the recording.
Translate the corrected text into the target language
For a browser workflow, Google's computer text-translation instructions document selecting the source and target languages, entering or pasting text, and copying the translated result. Use the corrected source transcript as that text input. This is a written-text operation; these instructions do not establish an audio-file upload feature.
Google currently limits a pasted segment to 5,000 characters. For a longer transcript, split at paragraph or sentence boundaries and keep enough surrounding wording to understand references such as “that agreement” or “she.” Record the segment order in your working document so no section disappears between pastes.
Copy each translated segment into the target document beside its source segment during review. Keep names and units easy to locate. Do not merge the translation into your source file, and do not delete an uncertainty note merely because the translated sentence reads smoothly.
Decide what the final document needs to preserve. A reader-facing summary may omit repetition; a full translated transcript has a different promise. Label the deliverable accordingly before shortening passages. Otherwise, a concise translation can be mistaken for a complete account of the recording.
Review the translation against the corrected source
Check the message across both documents, not just the target text's grammar. This original worksheet concentrates the review on details whose loss can change an instruction or attribution.
| Detail | Check in the corrected source | Check in the target text |
|---|---|---|
| Negation | Does the speaker say not, never, or an exception? | Is the prohibition or exception still present? |
| Quantities | What number, unit, date or price was actually said? | Does the same quantity and unit survive? |
| Names | Which person, place or product does this passage identify? | Can the reader identify the same subject? |
| Conditions | What must happen before the instruction applies? | Does the same condition still limit the instruction? |
When a passage fails this check, locate the source audio first. If the source words are wrong, correct that transcript and translate the passage again. If the source words are sound but the target wording changes the meaning, revise the translation against the source. That distinction keeps you from polishing the wrong sentence.
For a language you cannot read, you can check segment completeness, names and visible numbers, but those checks do not establish that the translation is accurate. A consequential passage needs review by someone who understands both languages. Keep unresolved wording marked rather than presenting fluency as certainty.
Finish with the original audio, corrected source and reviewed target document preserved separately. The useful result is a translation you can explain and correct: a reader can see which words came from the recording, which language change came afterward, and where uncertainty remains.
