User Profile
HeikoRa
MicrosoftJoined 7 years ago
User Widgets
Recent Discussions
We have updated Speech Studio
Speech Studio (https://speech.microsoft.com) has been updated to a new version. Besides enabling you to create custom models we have started adding ways to try out our out of the box capabilities. Check out the "Real-time Speech-to-text" area that let's you quickly test your audio without writing any code as well as our new Pronunciation Assessment feature that lets you evaluate a speakers fluency, accuracy and pronunciation which is great for language learning scenarios.Recent Blog posts for Speech Services
Here are some recent blog posts related to Speech Services: Enable read-aloud for your application with Azure neural TTS Add voice activation to your product with Custom Keyword Build an application that transcribes speech Azure Neural Text-to-Speech extended to support lip sync with visemeRe: Speech Model Limitations
Nathan Hess See the documentation here for recommended amounts of training data: Prepare data for Custom Speech - Speech service - Azure Cognitive Services | Microsoft Docs For audio training data the limit is 20h. For text data you could technically use more than the recommended data amounts, but typically gains would be minimal if any.Re: Learning/Training Different Accents
AlimaSam you can automate the training using our REST API see here: https://westus.dev.cognitive.microsoft.com/docs/services/speech-to-text-api-v3-0 You would need to create an endpoint for each user though which comes with some hosting cost.Re: Cost of Custom Neural Voice
AlimaSam we recently GA'ed the Custom Neural Voice creation capability. See docs here: Custom neural voice overview - Speech service - Azure Cognitive Services | Microsoft Docs Pricing information can be found here: Cognitive Speech Services Pricing | Microsoft Azure Model training is $52 per compute hour. Real-time synthesis: $24 per 1M characters Endpoint hosting: $4.04 per model per hour Long audio creation: $100 per 1M characters
