Showing posts with label N-gram. Show all posts
Showing posts with label N-gram. Show all posts

Thursday, November 01, 2012

Google Also Leads Voice Search: No Surprise

Image representing Google as depicted in Crunc...
Image via CrunchBase
When Siri came out there was talk finally Google has competition. I never bought into that. The front end might have been cute, but Apple and search? Come on.

Why I Won’t Be Using Google’s New iPhone Voice Search

For me it is less about moving from typing text to voice and more about moving from language to language to language.

That would be the faster way to wipe out illiteracy from the planet. If voice inputs and outputs can prove to be almost sufficient a lot of knowledge could move around right away.

Google offers up secret sauce on new voice search
this app gives Siri a run for her money. It is lightening fast, has a clean layout, and gives highly accurate results
Google explains how more data means better speech recognition
More data helps train smarter models, which can then better predict what someone say next ...... more data trumps better algorithms ..... For the voice search tests, the Google researchers used 230 billion words that came from “a random sample of anonymized queries from google.com that did not trigger spelling correction.” However, because people speak and write prose differently than they type searches, the YouTube models were fed data from transcriptions of news broadcasts and large web crawls. ...... As consumers demand ever smarter applications and more frictionless user experiences, every last piece of data and every decision about how to analyze it matters.
Large Scale Language Modeling in Automatic Speech Recognition
The n-gram approach to language modeling (predicting the next word based on the previous n-1 words) is particularly well-suited to such large amounts of data: it scales gracefully, and the non-parametric nature of the model allows it to grow with more data. For example, on Voice Search we were able to train and evaluate 5-gram language models consisting of 12 billion n-grams, built using large vocabularies (1 million words), and trained on as many as 230 billion words.
Enhanced by Zemanta