Implement text analysis solutions 4.2 Implement speech solutions
At an international event, attendees speak different languages and switch languages…
Study guide skill: Translate speech into other languages by using language models and Foundry Tools
Question B1-043
At an international event, attendees speak different languages and switch languages mid-conversation. You need low-latency speech-to-speech translation that doesn't require setting the input language and keeps the speaker's style and tone. Which option fits?
Select one answer.
Answer: C
Live Interpreter is part of real-time speech translation. It continuously identifies the spoken language without an input-language setting and delivers low-latency speech-to-speech translation in a natural voice that keeps the speaker's style and tone. The standard API expects a specified source language, and batch or document translation isn't real time.
Deep dive
How it works
Azure Speech supports real-time, multi-language speech-to-speech and speech-to-text translation of audio streams through the Speech SDK or Speech CLI. Interim results come back as speech is detected, and final results can be turned into synthesized speech. There are two ways to do real-time translation.
Live Interpreter is part of real-time speech translation and gives a fuller experience for conversations. It continuously identifies the language being spoken without needing an input-language setting, and it delivers low-latency speech-to-speech translation in a natural voice that preserves the speaker's style and tone. Its capabilities include:
- Unspecified input language: audio in a wide range of languages, with no expected language set in advance.
- Language switching: several spoken languages in one session, without restarting.
- Bring your own voice: personal voice for speech-to-speech translation.
Business meetings where participants speak different languages are one of the documented use cases. One current limit: transcription comes back in the target language, and source-language transcription isn't available yet.
The other options, and when you'd use them
- Standard speech translation with a fixed source language: The standard API is for sending audio in a specified source language and getting translated text, speech or both. It can also translate into two target languages in one call. Attendees switching languages mid-conversation break the fixed-source assumption.
- Custom Translator: This builds custom neural machine translation systems from parallel documents. It's for domain terminology, not live speech.
- Batch transcription then document translation: Batch transcription submits stored audio and returns results asynchronously, and document translation works on files. Neither is real time.
Exam tip
"Speakers switch languages" or "no input language set" plus "keep the speaker's voice and tone" means Live Interpreter. "Known source language" means standard speech translation.
References
Last reviewed: