Upload Cantonese audio or video and get a transcript that handles what makes Cantonese uniquely challenging: pervasive English code-mixing (我send咗個email畀你), a sentence-final particle system richer than any other Chinese variety, and the fundamental question of whether to output written Cantonese or standard Chinese.
Drop your file here or click to browse
.mp3, .wav, .m4a, .aac, .ogg, .flac, .mp4, .mov, .avi, .mkv, .webm·up to 500MB
Cantonese is not Mandarin with different pronunciation — it has its own grammar, vocabulary, and a written form that diverges significantly from standard Chinese. Hong Kong Cantonese speakers routinely mix English at the word level (我要book個room), use sentence-final particles (喎, 囉, 㗎, 嘅) that encode nuances no other Chinese variety expresses this way, and use characters like 嘅 (possessive), 咗 (completed), and 唔 (negation) that don't exist in Mandarin. Vocova's AI is built for Cantonese specifically, handling these features natively rather than forcing Cantonese speech through a Mandarin-shaped pipeline.
Drag and drop or select a file containing Cantonese speech. Handles everything from Hong Kong business calls to TVB dramas to diaspora community recordings.
The engine recognizes Cantonese-specific grammar and vocabulary, preserves sentence-final particles, and handles English-Cantonese code-mixing at the word and phrase level.
Review the Cantonese transcript, edit inline if needed, and export in your preferred format.
Hong Kong Cantonese doesn't just borrow English words — it integrates them into Cantonese grammar. Speakers say 我send咗個email畀你 (I sent you an email) with Cantonese aspect markers wrapping English verbs. The AI preserves this code-mixing pattern faithfully, outputting English words where the speaker used them rather than attempting to translate everything into Chinese characters.
Cantonese has over 30 sentence-final particles that encode mood, attitude, and social meaning: 㗎 (assertion/emphasis), 喎 (surprise/new information), 囉 (obviousness), 嘛 (it should be clear), 啦 (suggestion/urging). These aren't optional — dropping them changes the meaning. The AI captures the particle actually spoken rather than omitting them or substituting a generic marker.
Written Cantonese uses characters that standard Chinese does not: 嘅 (possessive, = Mandarin 的), 咗 (completed action, = Mandarin 了), 唔 (negation, = Mandarin 不), 冇 (not have, = Mandarin 没), 嘢 (thing, = Mandarin 东西), 佢 (he/she, = Mandarin 他/她). The AI uses these characters naturally because it processes Cantonese as its own language system.
Cantonese has 6 citation tones — more than Mandarin's 4 — plus changed tones that function as morphology (e.g., the high-rising changed tone marks familiarity or diminutive meaning). The AI uses this tonal information alongside context to select the correct characters from Cantonese's rich tonal landscape.
The AI understands Hong Kong-specific vocabulary (巴士 for bus, 的士 for taxi, 冷氣 for air conditioning), institutional terms (立法會, 廉政公署), and the characteristic speech rhythm of Hong Kong Cantonese. It also handles the Guangzhou Cantonese variety with its own vocabulary preferences.
Hong Kong business Cantonese mixes English terminology freely — 我哋要review下個proposal先 is standard office speech. The transcript captures this naturally so your meeting minutes reflect how the conversation actually happened.
Export as SRT or VTT for Hong Kong dramas, YouTube content, and social media videos. The subtitles include Cantonese-specific characters and code-mixed English, matching how Cantonese audiences expect to read their language.
Turn Cantonese news broadcasts, radio programs, and talk shows into searchable text for media monitoring, journalism, and content analysis across Hong Kong's media landscape.
Cantonese oral histories, community recordings, and cultural content deserve transcription in written Cantonese — not a Mandarin approximation. The AI produces text that reflects how Cantonese is actually spoken, preserving linguistic features that standard Chinese transcription would erase.
Capture Cantonese meetings and calls with English code-mixing preserved. Get written records that reflect the actual language of Hong Kong business communication.
Generate subtitles and written content from Cantonese videos and podcasts using written Cantonese characters (嘅, 咗, 唔) that your Hong Kong audience recognizes.
Get transcripts that preserve sentence-final particles, code-mixing patterns, and Cantonese-specific grammar for linguistic analysis and sociolinguistic research.
Transcribe family recordings, community media, and cultural content in written Cantonese for language preservation across generations in North America, Europe, and Southeast Asia.
Vocova outputs written Cantonese by default — using Cantonese-specific characters (嘅, 咗, 唔, 冇, 佢) and Cantonese grammar patterns. This reflects how the speaker actually talks. Standard written Chinese (書面語) would replace these with Mandarin equivalents (的, 了, 不, 没, 他), which is a translation step, not a transcription.
The AI preserves code-mixing as spoken. When a speaker says 我send咗個email畀你, the transcript shows exactly that — English words embedded in Cantonese grammar. It handles word-level mixing (individual English nouns and verbs), phrase-level switching (entire clauses in English), and even hybrid constructions where Cantonese affixes attach to English words.
Yes. These characters are fundamental to written Cantonese: 嘅 (possessive marker), 咗 (perfective aspect), 唔 (negation), 冇 (not have), 嘢 (thing), 畀 (give), 佢 (third person pronoun). The AI uses them naturally because it processes Cantonese as its own language system, not as Mandarin with substitutions.
Cantonese has 6 citation tones versus Mandarin's 4, plus a system of changed tones that function as word-formation. The high-rising changed tone, for example, can mark diminutive meaning or familiarity. The AI's acoustic model is trained on Cantonese tonal patterns specifically, so it doesn't confuse Cantonese tone distinctions with Mandarin categories.
Yes. Cantonese has over 30 sentence-final particles — 㗎 (emphatic assertion), 喎 (contrary to expectation), 囉 (it's obvious), 嘛 (should be clear), 啦 (urging), 吖 (friendly suggestion), 咩 (rhetorical question), and many more. These carry critical pragmatic meaning and the AI preserves the specific particle used.
Yes. While Hong Kong Cantonese is the primary training focus (and includes heavy English code-mixing), Guangzhou Cantonese is also well-supported. The AI recognizes vocabulary differences between the two — for instance, Guangzhou speakers may use more standard Chinese vocabulary while maintaining Cantonese phonology and grammar.
In some contexts (particularly in Guangdong or among younger Hong Kong speakers), Cantonese and Mandarin may mix. The AI can handle this, though its primary strength is Cantonese-English code-mixing which is far more common in everyday Hong Kong speech. Cantonese-Mandarin mixing is transcribed with best-effort character selection.
Yes. Expressions like 唔該 (thank you/excuse me), 點解 (why), 乜嘢 (what), 幾時 (when), and 邊度 (where) are output in their Cantonese forms rather than being normalized to Mandarin equivalents (謝謝, 為什麼, 什麼, 什麼時候, 哪裡). This is essential for a transcript that actually represents Cantonese speech.
Transcribe Chinese (Mandarin) audio and video with AI
Transcribe Japanese audio and video with AI
Transcribe Korean audio and video with AI
Upload any audio file and get accurate text instantly
Upload audio in any language and translate it to 140+ languages
Upload audio or video and get ready-to-use subtitle files
Upload a file or paste a link from YouTube, podcasts, cloud storage, and 1,000+ platforms. Get an accurate transcript in minutes. No credit card required.