Vocco / Guides

Transcribe audio to text on a Mac, offline

Drop a recording into Vocco and get a transcript with timestamps and speaker labels, made on your Mac. This page covers the files it reads, what the export looks like, how it does on speech that mixes Chinese and English, and when another app is the better pick.

Updated October 28, 2026 · Written by the developer of Vocco

What you can transcribe

• Audio files: MP3, WAV and M4A.
• Video files: MP4 and MOV. Vocco reads the audio track.
• Several at once. Drop in a batch or click Add files. Files queue up and run one at a time, each with its own progress bar.
• Long files. The free version transcribes files up to 15 minutes, without speaker labels. With Pro, the drop zone says up to 4 hours per file. Speaker labels are skipped on files longer than 3 hours; you still get the transcript.
• Live recordings. The Live transcription view writes text while you talk. When you stop, Pro labels the speakers (for recordings up to 90 minutes) and, unless you turn it off, keeps the audio so you can play it back.

Three steps

1. Install Vocco and open it while online: it downloads Qwen3 ASR (987 MB) in the background and uses it from then on. There is nothing to sign up for.
2. Open File transcription and drop your files in. The engine button above the queue sets which engine the files use.
3. When a file shows Done, click Open to read it, or Export SRT right from the queue. Inside a transcript you can also Copy all or Export TXT.

Speaker separation is a Pro feature, included in every new install's first 14 days, and on by default. The switch is in Settings → Transcription.

Known issue (still in 1.0.8): on long recordings, one person can be split into several speaker numbers (a 17.5-minute meeting with four people came back with 24). A fix is coming. For long meetings, turn speaker separation off for now; it also makes the file faster.

What the transcript looks like

Each paragraph has a start time, plus a speaker number when Vocco hears more than one voice. In the TXT export, every paragraph starts with its time in brackets and the speaker name. The SRT export turns the same paragraphs into subtitle cues, with the speaker name at the start of each cue.

Two things to know before you use the SRT for subtitles. Times are whole seconds. And with speaker separation on, one cue is one speaker's whole turn, which can be long. For shorter cues on a multi-speaker recording, turn speaker separation off first; Vocco then breaks the text at pauses instead.

Accuracy on mixed Chinese and English

On 300 utterances of real Mandarin–English conversation from the public ASCEND corpus, Qwen3 ASR had the lowest mixed error rate: 0.157, about 16 errors per 100 characters and words, with 29% of English words wrong. Apple's on-device recognizer got 53% of English words wrong. Whisper large-v3-turbo scored 0.277 with automatic language detection and 0.373 pinned to Chinese, which is how Vocco runs it under a Chinese interface.

Paraformer, also in Vocco, scored 0.170 overall but did better on the 62 sentences that actually switch language (0.145 against 0.170). The gap between the two is within noise, so for mostly-Mandarin recordings, try both.

All 300 utterances are short clips from one conversation between two speakers. That is enough to rank the engines against each other, not to predict the exact error rate on your recording. Method and per-utterance output: benchmark.

Vocco, MacWhisper and Voice Memos

I make Vocco. MacWhisper details are quoted from macwhisper.com, its Gumroad page and its help center; Voice Memos details from Apple's user guide. All were checked on October 26, 2026.

MacWhisper is the better pick if you transcribe a lot of English or single-language files and want more around it: a built-in transcript editor, renaming and merging speakers, watched folders, YouTube links, meeting recording from Zoom and Teams and more export formats.

Voice Memos is enough if you already record with it on an Apple-silicon Mac with macOS Sequoia or later, and only need searchable text you can copy. Apple notes that transcription isn't available in every country or region.

Vocco fits when recordings mix Chinese and English, when you want speaker labels for a one-time $29 at launch ($39 after) rather than €64, or when the audio should not leave the Mac at all.

When not to use Vocco

• You need VTT, Word files or frame-accurate subtitles. Vocco exports TXT and SRT, and SRT times are whole seconds.
• You want to edit the text or name the speakers inside the app. Vocco 1.0.8 has no transcript editor, and speakers are numbered, not named. Export TXT and edit it elsewhere.
• You want to capture a Zoom or Teams call directly. Vocco records from input devices such as your microphone, not from your Mac's audio output. Save the meeting recording as a file, then transcribe that.
• Most of your audio is in another language. Vocco is built and tested for Mandarin and English; Japanese, Korean and French were only spot-checked, one language at a time. Chinese comes out in Simplified characters only, and we have not tested Cantonese–English mixing.
• You use an Intel Mac or macOS 13. Vocco needs Apple silicon and macOS 14 or later.

Questions

Is my audio uploaded anywhere?

No. The downloadable engines and speaker separation run on your Mac, and Vocco has no server that receives audio. Apple Speech is used on-device only: if macOS cannot recognize the language on this Mac, Vocco stops with an error instead of letting macOS send the audio to Apple.

Can it transcribe a recording that switches between Chinese and English?

Yes, with Qwen3 ASR, the default download. It works out the language as it goes, so one file can mix both. Whisper in Vocco is fixed to one language per session, so don't use it for these recordings.

How long does a one-hour recording take?

On an M5 Pro, about 6 minutes with Qwen3 ASR, or about 11 minutes with speaker labels on. We timed the 1.0.5 code on an hour of real Mandarin–English conversation. Macs with fewer CPU cores take longer.

Can I rename speakers or tell it how many there are?

Not yet. Speakers are numbered in the order they first talk, and Vocco estimates how many there are by itself. On long recordings the current version tends to find too many: one person can show up as several speakers.

Does transcription need an internet connection?

No. You need to be online once to download an engine (Qwen3 ASR is 987 MB). No sign-in is needed; when the Mac is online, the app also checks its trial or license. After that, file transcription works with Wi-Fi off.

How this page was made: formats, export behaviour, limits and settings come from Vocco 1.0.5's code; timings and speaker counts come from our test runs of that code on October 26–27, 2026; accuracy comes from our published benchmark; MacWhisper and Voice Memos details are quoted from their official pages on October 26, 2026. Drafted with AI assistance and checked against those sources.