Voice activity detection
VAD
Related terms
Text normalization
Text normalization is the step where a voice agent converts written text into the words it should actually speak, so numbers, currency, dates, abbreviations, and email addresses sound the way a person would say them rather than being read character by character. For example, $1,250.00 becomes "one thousand two hundred fifty dollars," 03/04 is read as the correct date for the caller's locale, and an email address is spoken as grouped characters with "at" and "dot." Skipping this step is a common cause of callers asking for repeats or escalating to a human agent partway through a call. OnClarity's AI agents work across chat and voice; on voice, text normalization is what lets spoken confirmations, such as order numbers or account details, read back clearly on the first try. Related: voice activity detection.
Call transcription
Call transcription converts recorded or live call audio into searchable text using automatic speech recognition (ASR). It can run in real time, powering live agent guidance, or in batch after the call ends. Batch processing generally yields higher accuracy. Quality transcripts include speaker diarization, timestamps, and redaction of sensitive data like card numbers. Accuracy is measured by word error rate. OnClarity’s AI Agent QA uses transcripts to score 100% of conversations for QA and compliance. Example: a team searches six months of transcripts for every mention of a failed payment flow to find a root cause.
Voice AI
Voice AI is automation that handles spoken conversations using speech recognition, language understanding, and speech synthesis. It carries constraints text automation does not: recognition errors on names and numbers, sensitivity to accents and dialects, and a much lower customer tolerance for response latency.
Latency
Latency is the delay between a customer’s message and the first useful response. In live chat and voice it directly shapes perceived quality, and it is one of the few AI quality measures a customer notices immediately, well before judging whether the answer was correct.
