Turning Written Chinese Scripts into Spoken Hong Kong Cantonese: 12 Giveaways of AI Read-Aloud and How to Fix Them
When a voiceover script is written in formal Chinese, even an accurate AI voice sounds like a news bulletin. This CantoneseTTS.com Studio Benchmark lists 12 common giveaways, from formal pronouns to currency phrasing, each with a before and after rewrite, and shows a 53-character written script becoming 30 characters of natural spoken Hong Kong Cantonese.
按代詞、動詞時態、句式、語氣助詞同中英夾雜整理
第 4 節完整示範段(不計標點),讀出嚟約短 5 秒
代詞先行、動詞句式次之、最後補語氣助詞
To convert written Chinese into spoken Hong Kong Cantonese for voiceover, work in three passes: swap pronouns (您 to 你, 我們 to 我哋), swap verbs and sentence patterns (沒有 to 冇, 了 to 咗), then add sentence-final particles such as 啦, 喎 and 㗎. In our benchmark a 53-character script shrank to 30.
Compare Flat Reading vs. Directed Cantonese Emotion
「朝早返工前去茶記,一份餐蛋麵加凍檸茶少甜,三十八蚊搞掂!老闆仲會記得你飲開咩,呢種人情味真係其他地方搵唔到㗎!」
Giveaways 1 to 3: Pronouns and Structural Particles
第 1 組:「您」改「你」。「您」喺廣東話口語幾乎冇人講,AI 照讀出嚟會變成怪腔。原稿(書面語):「感謝您的支持。」執靚後(地道香港口語):「多謝你支持!」第 2 組:「我們、你們、他們」改「我哋、你哋、佢哋」。原稿:「他們都說好用。」執靚後:「佢哋個個都話好用。」
第 3 組:「的」改「嘅」,或者直接刪走。原稿:「這是我們最受歡迎的產品。」執靚後:「呢款係我哋最多人買嘅。」留意口語好多時連「產品」都唔使講,用量詞「款、隻、件」頂替,讀落更似真人。喺 CantoneseTTS.com 團隊內部實測基準案例(Studio Benchmark)嘅執稿流程入面,代詞類改動最簡單、見效最快,所以放喺第一步。
- 您 → 你;我們、他們 → 我哋、佢哋
- 的 → 嘅,或者用量詞(款、隻、件)代替名詞
- 這、那 → 呢、嗰,例如「這款」改「呢款」
Giveaways 4 to 6: Copulas, Negation and Aspect Markers
第 4 組:「是」改「係」。原稿(書面語):「這是限量版。」執靚後(地道香港口語):「呢個係限量版嚟㗎!」第 5 組:「沒有、不」改「冇、唔」。原稿:「我們沒有分店,不設送貨。」執靚後:「我哋冇分店,亦唔送貨。」留意「不」唔可以一律改「唔」,「不過」「不得了」呢類固定詞要保留。
第 6 組:「了」改「咗」,「正在」改「緊」。原稿:「新貨已經到了,我們正在安排出貨。」執靚後:「新貨到咗喇,我哋而家執緊貨。」呢組最考功夫,因為書面語「了」有時表示完成,有時只係語氣,要按意思揀「咗」定「喇」。用香港語言學學會粵拼方案對照讀音,例如「咗」讀 zo2、「喇」讀 laa3,可以幫你確認引擎讀得啱唔啱。
- 是 → 係;沒有 → 冇;不 → 唔(固定詞除外)
- 已經……了 → ……咗;正在 → ……緊
- 表示完成用「咗」,報告新狀況用「喇」
Giveaways 7 to 9: Sentence Order, Formal Courtesy and Connectives
第 7 組:「給」改「俾」,同時調語序。原稿(書面語):「我們會給你一份禮物。」執靚後(地道香港口語):「我哋會送份禮物俾你。」廣東話習慣將「俾你」放喺句尾。第 8 組:公文式客套話。原稿:「如有查詢,歡迎與我們聯絡。」執靚後:「有咩想問,隨時搵我哋!」
第 9 組:長句連接詞。「因此、然而、此外、並且」喺口語會變成「所以、不過、仲有、同埋」。原稿:「此產品輕巧便攜,並且續航力強,因此深受上班族歡迎。」執靚後:「呢部機好輕好易帶,電又夠長氣,所以好多返工一族都鍾意。」執靚後拆成三段短句,每段之間有自然停頓位,AI 讀出嚟換氣更似真人。
- 給你 X → 送 X 俾你,「俾你」放句尾
- 歡迎與我們聯絡 → 隨時搵我哋
- 因此、然而、此外 → 所以、不過、仲有,並拆成短句
Giveaways 10 to 12: Particles, Code-Mixing and Numbers, with a Full Rewrite
第 10 組:句尾補語氣助詞。「啦」帶催促同親切,「喎」帶提醒,「㗎」帶肯定,冇助詞嘅句子 AI 讀出嚟好平。第 11 組:硬譯英文詞。香港人講「搞 promotion」「check 吓」「book 位」,寫成「促銷活動」「檢查一下」「預約座位」反而唔自然,呢啲係 Konglish 夾雜要保留嘅位。第 12 組:數字同貨幣。「港幣二百四十元」改「二百四蚊」,「百分之二十折扣」改「八折」。
完整示範(CantoneseTTS.com 團隊內部實測基準案例)。原稿(書面語):「如果您正在尋找一份適合送給母親的禮物,我們為您推介這款香薰蠟燭。現在購買,即可享有八折優惠,價格只需港幣二百四十元。」執靚後(地道香港口語):「搵緊禮物送俾阿媽?呢款香薰蠟燭啱晒你!而家買有八折,只係二百四蚊咋!」不計標點,字數由 53 字減到 30 字,用到第 1、3、6、7、10、12 組公式。以每秒約 4 至 5 字計,讀出嚟短咗大約 5 秒,用 cantonese-kol-boy 配開心情緒生成,語氣似朋友推介多過廣告讀稿。
- 句尾助詞:啦(催促)、喎(提醒)、㗎(肯定)、咋(強調少)
- 香港人慣講嘅英文詞保留原文,例如 promotion、check、book
- 貨幣同折扣:港幣二百四十元 → 二百四蚊;百分之二十折扣 → 八折
Frequently Asked Questions
Q1. Is there a tool that converts written Chinese to spoken Cantonese automatically?
AI can produce a first draft, but choices such as 了 to 咗 or 喇, and whether 不 becomes 唔, still need human review. Use the 12-point checklist in this article to check each sentence.
Q2. Can AI pronounce Cantonese colloquial characters correctly?
Mainstream Cantonese TTS handles common colloquial characters such as 咗, 嘅, 啲 and 冇. For rare characters or words with multiple readings, check the LSHK Jyutping scheme and confirm by listening.
Q3. Should formal voiceovers also be converted to spoken Cantonese?
News, legal notices and government announcements can stay in written Chinese with a neutral tone. Ads, social videos and product explainers should be converted, because viewers expect everyday conversational delivery.