A telecom quality team has 50,000 recorded support calls to transcribe overnight. The calls contain heavy use of network equipment model names that generic recognition often gets wrong. Which approach best meets these requirements?
Show answer and explanation
Answer: D
Batch transcription handles large volumes of recorded audio asynchronously, which fits an overnight deadline. Custom Speech lets the recognizer learn the equipment model names from domain training data, improving accuracy on terms that generic models miss. Real-time recognition, language understanding applied to raw audio, and text glossaries do not solve the combined volume and vocabulary problem.
Why the other options are wrong
- A. Real-time recognition is intended for live audio with low latency. SSML affects synthesized speech output, so slowing text output does nothing to improve transcription of the calls.
- B. Conversational language understanding works on text utterances to predict intents. It cannot transcribe raw audio, so it would not produce the transcripts the team needs.
- C. A Translator glossary standardizes term rendering during translation. It is applied to text in another language and does not improve how the speech engine recognizes model names.