Google states that they have built even better neural network acoustic models using Connectionist Temporal Classification (CTC) and sequence discriminative training techniques. To us regular folk who are not engineers, whenever you perform a voice search on Android, your results will be more reliable.
Thanks to Google’s work, the new models will provide more accurate results, are more robust to noise, and will provide faster response times to voice search queries. These changes are now available to users of the service, so the only thing you need to do is try it out. Google does the heavy lifting.
Inside Google’s blog post announcing these changes, they speak of exactly how they made this possible. For those curious about sound waves and audio technology, give it a read.