WhatsApp and artificial intelligence
Speech-to-text
Speech-to-text, or audio transcription, is the automatic conversion of spoken words into written text, also known as speech recognition. On WhatsApp it lets an assistant understand a voice note as if the person had typed it, with the risk of mishearing a number or a name.
Also known asaudio transcription · speech recognition · voice-to-text · voice note transcription
How it is used in a residence
Many people would rather talk than type, especially in a hurry or with their hands full: "hey, there's been no hot water in 312 since this morning". An assistant that only reads text draws a blank at that voice note. With transcription, the audio becomes text and from then on it is handled like a written message: it is understood, classified and, if it is a fault, turned into a maintenance ticket.
Quality is measured by the word error rate, and it gets worse with background noise, echo, accents and unusual vocabulary. In a residence the errors that matter are small and costly: a room, a time, a meal. "312" heard as "3012" opens the ticket for a room that does not exist, and a misheard "I'm not having dinner" changes someone's meal. A voice note in another language raises a further question, that of language detection.
Worth knowing
What people take for granted: that a voice note is always understood. There are size limits, recordings full of background noise and systems that only transcribe with one particular model. When a recording cannot be understood, the right thing is to say so and ask for the message in writing, not to answer what the system thinks it heard.
Sources
Related terms
- AI assistantAn AI assistant is a program that holds a conversation in natural language and, with a language model behind it, understands what it is told even without keywords. In a residence it handles the messaging channel: it answers what is in the house rules, takes requests and passes the rest to a person.
- Language detectionLanguage detection is the task of working out automatically which language a text is written in. In a WhatsApp assistant it decides which language to reply in, and what it finds hardest is telling closely related languages apart or handling sentences that mix two languages.