Unbabel Talk: Difference between revisions
From HLT@INESC-ID
m (Update youtube link) |
m (Add link section) |
||
Line 16: | Line 16: | ||
voice morphing and transcription correction through Unbabel’s | voice morphing and transcription correction through Unbabel’s | ||
community. | community. | ||
== Links == | |||
[https://www.inesc-id.pt/publications/14969/pdf/ Paper link] | [https://www.inesc-id.pt/publications/14969/pdf/ Paper link] | ||
[https://www.youtube.com/watch?v=cF2a3BRgo7Q Youtube link] | [https://www.youtube.com/watch?v=cF2a3BRgo7Q Youtube link] |
Revision as of 16:14, 27 November 2020
Abstract
Unbabel Talk is a speech-to-speech translation application that provides human certified translations for voice instant mes- saging (IM) in multilingual scenarios. By combining Unba- bel’s translation pipeline with state-of-the-art automatic speech recognition (ASR) and text-to-speech (TTS) models, Unbabel Talk can be used to send a voice message in a language of choice through popular messaging platforms. The app further ensures that translations have high quality, either by certify- ing them through Unbabel’s own quality estimation (QE) tool and/or through Unbabel’s community of translators. There are two versions of the app. On version 1, the app synthesizes audio that can be delivered with male or female standard voices. Ver- sion 2 has features that are currently being developed, such as voice morphing and transcription correction through Unbabel’s community.