Netreo is now BMC. Read theBlog

Azure Cognitive Services: Custom Speech As An Innovative Machine Learning Approach For Speech Recognition

By: Netreo
August 10, 2018

The latest and newest crossing point driving ambiance is the Voice. The speech recognition has been really helpful in the last few years as it is becoming our daily life digital assistant. We use it daily writing emails and other necessary documents. Apart from this, it also helps in dictating the transcriptions of the lectures and meetings. These situations are possible mainly because of the hard work and several months of studies and research in speech identification and technical jumps made possible with the help of neural networks. Microsoft is at the top of the list mainly because it is at the front position of speech recognition with all the research and study results.

Cognitive Services Speech API provides the developers with the new art speech models. Exceptional situations, which use the domain-specific vocabulary or multifaceted acoustic conditions, also offer Custom Speech Service which allows the manufacturers to inevitably tune speech recognition models to their particular needs. These services have been previewed on a wide range of scenarios with customers.

Custom speech service

Speech recognition services contain several components. Some components include the acoustic and language models. If an application consists of vocabulary items that do not frequently occur in our daily life communication, then customizing the language model will significantly improve the recognition correctness. Users can easily upload the documented or word-based data like sentences and words of the focused domain in order to build the language models that can be deployed and retrieved through the Speech API.

One of the most common and typical examples is that of the university lectures as they usually consist of domain-specific terms. For instance, in a biology lecture, we usually hear the terms and words like “Nerodia erythrogaster” which are quite specific and important to transcribe properly. The Presentation Translator, an add-in to the PowerPoint is used to modify the language model centered on the slide content, reports the lecture or any other presentation situation providing extremely precise transcription results for any domain-specific audio.

Customization sample - Azure Cognitive Services: Custom Speech As An Innovative Machine Learning Approach For Speech Recognition

Likewise, altering and modifying the acoustic model also allows the speech recognition model to be more precise in specific situations. Let’s say if a voice-enabled application is intended to be used in a factory, a practice acoustic model can correctly identify speech in the presence of loud or persistent background noise.

Speech recognition customization workflow - Azure Cognitive Services: Custom Speech As An Innovative Machine Learning Approach For Speech Recognition

Custom Speech service allows acoustic and language model adaptation with no coding. User interface directs allows for data import, model variation, and evaluation by measuring the word error rate. It also guides through model deployment at scale, so models can be accessed by an application on any devices.

Further reading

Find out how Netreo can help with all Azure monitoring needs. Request a Demo Today!

Ready to get started?

Get in touch or schedule a demo

Get Started Learn More