Microsoft releases upgrades to Azure AI Speech at Build 2024

At its annual Construct developer convention on Tuesday, Microsoft introduced new options for its Azure AI Speech service that improve voice-enabled, generative AI-powered app growth.

Azure AI Speech is already getting used for “quite a lot of use circumstances together with name analytics (audio, textual content), medical transcription (audio, imaginative and prescient, textual content), captioning (audio/video, transcription, translation) and chatbots (audio, GPT),” Microsoft stated within the launch. The service has quite a few capabilities thus far, together with changing audio into textual content captions for a broadcast or extracting the addresses talked about on a cellphone name. 

One spotlight of OpenAI’s GPT-4o reveal final week was an improved Voice Mode, which targeted on the improved high quality of the voice given to this system’s responses. Operating to maintain up, Microsoft introduced it’s making Private Voice typically out there. 

The function lets customers “create and use their very own AI voices for numerous functions, similar to voice assistants, speech translation, and video content material creation,” the discharge defined. 

One other new functionality is speech analytics, now out there in preview. Accessible inside Azure AI Studio, Adobe’s growth setting, it’s supposed to handle what the corporate calls the “comfortable” evaluation of cellphone calls or different audio sources. A comfortable aspect of a name could possibly be semantic evaluation, or how the caller appears to really feel, which is presumably subtler than the content material of the decision itself.

Semantic evaluation may detect particulars just like the “diploma of empathy proven, dedication of the individuals and energy of the arguments made and even predict potential dialog flows,” the discharge explains. 

In a transcript of a name, for instance, sections could possibly be labelled with a ranking of every speaker’s phrase as “constructive,” “unfavorable” or “impartial.” You’ll be able to take a look at an interactive demo right here. 

To make fast evaluation potential, Microsoft can also be rolling out Quick Transcription, which the corporate claims is “a sport changer for transcription at massive” as a result of “it might now transcribe 40x quicker than real-time (real-time issue<1).” 

In line with the corporate, Quick Transcription can save name heart brokers “hundreds of hours” by eliminating the necessity to manually take notes on a name, and medical doctors and nurses can analyze conversations with sufferers in seconds. “Media and content material creators can analyze and extract insights from podcasts or interviews as quickly as they full,” the discharge continued.

Microsoft stated the function might be made out there subsequent month. 

To satisfy the necessity for disseminating content material globally, Microsoft additionally teased automated video dubbing, which interprets content material, synthesizes a voice within the goal language, and syncs it to the video of the speaker. 

Moreover, the corporate introduced updates to its multi-lingual translation function, similar to the flexibility to change languages for captioning whereas an individual is watching a broadcast.