- Speech-to-Text (STT) adds language support for UK English and Arabic, and¬†Text-to-Speech (TTS) now supports Brazilian Portuguese.
- Our STT service can now return a much richer set of confidence-based annotated results, which makes it easier for downstream processing to select the best interpretation of what was said. For example, in the utterance “I move it very often”, the phrase “move it” may be misrecognized as “move”. Our new system now returns word-annotated confidence scores for multiple hypotheses as shown in the picture, so the phrase “move it” can be located among the alternative hypotheses and also assigned an overall confidence of .2145 (the product of the individual word confidences).
- We have also added the capability to leverage these expanded results to find user-specified key words and phrases in the audio stream.¬† A user can now specify one or more words to search for in the audio stream and the service will return the location(s) in the audio stream of where the keyword(s) are found as well as their confidence.
- Developers now have the ability to alter the way specific words are spoken by the TTS service. For example, the pronunciation of¬† the word “tomato” can be altered to sound like “tomahto”.¬†
STT Documentation | Demo¬†¬†¬†¬†¬†
TTS Documentation | Demo