很多人是為了語音記帳才下載哈啦Money。這篇說明怎麼講辨識率最好、辨識是誰做的,以及怎麼從「記得起來」走到「省得下來」。
Voice entry is why most people download SpendSay in the first place. Here's how to phrase an entry so it's recognised, who does the recognition, and how to get from tracking to actually spending less.
記帳最累的不是記,是每一筆都要打開 App、選分類、打金額、打品名。多數人不是不想記帳,是懶得走完那四個步驟,於是拖到週末,週末就忘了買過什麼。
語音記帳把這四步壓成一句話:點麥克風,說出你剛剛花了什麼、多少錢,App 辨識出金額,並自動歸到對應的分類。記帳最好的時機是消費當下,因為只有那一刻你還記得細節,而語音把「當下就記完」的成本降到說一句話。
語音是我們的起點,不是全部。發票掃描、消費明細匯入、手動輸入都在,挑最順手的用就好。
最常見的抱怨是「有時候要講四五次才對,那還不如用打的」。多數情況下,問題出在話講得太短。
| 你說的話 | 結果 |
|---|---|
| 米漿 10 元 | 字太少,容易辨識錯 |
| 一杯米漿 10 元 | 好很多 |
| 買一杯米漿 10 元 | 更穩定 |
原因是:語音辨識靠上下文猜。字越少,能參考的線索越少,同音字就會亂跳。加上量詞(一杯、一份、一張)、動詞(買、付、繳)和單位(元),辨識引擎就有東西可以判斷,準確率會明顯提高。
順帶一提,多講幾個字對日後查帳也有好處:品名越具體,幾年後用關鍵字越找得到。萬一還是辨識錯了,點一下那筆就能改。
語音轉文字這一段是手機系統做的:Android 用 Google 的語音辨識,iOS 用 Apple 的。所以辨識品質很大一部分取決於你手機的系統設定與當下的網路狀況;如果一直辨識成別的語言,先檢查手機的語音輸入語言設定。
轉成文字之後,才輪到我們判斷金額與分類。我們不會儲存你的錄音。
有人問我們,記帳記了半天,到底怎麼變成真的省下錢?三個步驟,順序不能顛倒。
至少一個月,能撐到三個月更好。重點是完整,不是漂亮。四種方式混著用,哪個當下最快就用哪個:
資料不完整,後面兩步都是假的。這也是為什麼我們花這麼多力氣讓記帳變快,快才記得完。
累積了一兩個月之後,看兩張報表就夠了:
找到那一類之後,直接給它設一個月預算。預算可以設整體,也可以只設某個分類、子分類,或某個標籤(例如只給「手搖飲」或「沖繩旅行」設)。設好之後,記帳頁就會顯示本月預算的進度,越接近上限越明顯。
真正讓人省下錢的,不是月底看到超支的報表,而是月中記帳時看到「這個月已經用掉八成」的那一眼。
記消費日還是付款日、信用卡怎麼分類、標籤怎麼用,看記帳觀念指南;電子發票與手機條碼,看電子發票指南。其他問題歡迎到常見問題,或到我們的 Facebook 粉絲專頁私訊小編。
The tiring part of tracking spending isn't the tracking. It's opening the app, picking a category, typing an amount and typing a name, four steps, every single time. Most people don't dislike keeping a ledger; they put it off until the weekend, and by the weekend they've forgotten what they bought.
Voice entry collapses those four steps into one sentence: tap the microphone, say what you spent and how much, and the app picks out the amount and files it under the right category. The best moment to record an entry is right after you spend, because that's the only moment you still remember the details, and voice brings the cost of recording it then down to one sentence.
Voice is where we started, not all there is. Invoice scanning, CSV import and manual entry are all there; use whichever is fastest at the time.
The most common complaint is "sometimes I have to repeat myself four or five times, I might as well type it". Nearly always, the problem is that the sentence was too short.
| What you say | Result |
|---|---|
| Rice milk 10 | Too few words, easily misheard |
| A cup of rice milk, 10 dollars | Much better |
| Bought a cup of rice milk for 10 dollars | More reliable still |
Speech recognition works from context. The fewer words it hears, the less it has to go on, and similar-sounding words start winning. Add a counter word, a verb (bought, paid) and a unit, and the engine has something to reason with. Accuracy goes up noticeably.
Saying a few more words pays off later too: the more specific the item name, the easier it is to find by keyword years from now. And if something is still misheard, one tap edits the entry.
The speech-to-text step is done by your phone's own system: Google's recogniser on Android, Apple's on iOS. So quality depends a lot on your phone's settings and your connection at that moment; if entries keep coming out in the wrong language, check your phone's voice input language first.
Only after it becomes text do we work out the amount and the category. We never store your audio.
People ask us how tracking turns into money saved. Three steps, and the order matters.
One month minimum, three is better. What matters is that it's complete, not that it's tidy. Mix the four input methods and use whichever is quickest in the moment:
If the data isn't complete, the next two steps are fiction. That's why we put so much work into making entry fast: fast is what makes it complete.
After a month or two, two reports are enough:
Once you've found it, give it a monthly budget. A budget can cover everything, or just one category, one subcategory, or one tag (bubble tea, say, or the Okinawa trip). From then on your ledger screen shows this month's progress against it, and it gets harder to ignore as you approach the limit.
What actually saves money isn't the over-budget report at the end of the month. It's glancing at "80% used" halfway through, while you're recording something else.
For which date to log, where a credit-card bill goes, and how to use tags, read the guide to keeping a ledger. For e-invoices and the carrier barcode, read the e-invoice guide. Anything else, see the FAQ or message us on our Facebook page.