This paper describes recent experiments using the HTK large vocabulary continuous speech recognition system [6].The system uses state-clustered mixture Gaussian cross-word triphone HMMs to allow an appropriate balance of acoustic modelling detail (model complexity) and parameter estimation accuracy for a given training corpus.The system decoder is able to integrate cross-word triphone acoustic models and trigram language models into a single recognition pass.The approach was evaluated in the ARPA November 1993 Wall Street Journal (WSJ) evaluation with both 5k word and 20k word vocabularies.The HTK based system had the lowest error rate reported on three of the four tests entered and the second lowest error rate on the fourth test.Recently, our focus has been to optimise and further enhance the capabilities of the system.However

The present disclosure is directed to input suggestion. In particular, the methods and systems of the present disclosure can: receive, from a first application executed by one or more computing devices, data indicating information that has been presented by and/or input into the first application; generate, based at least in part on the received data, one or more suggested candidate inputs for a second application executed by the computing device(s); provide, in association with the second application, an interface comprising one or more options to select at least one suggested candidate input of the suggested candidate input(s); and responsive to receiving data indicating a selection of a particular suggested candidate input of the suggested candidate input(s) via the interface, communicate, to the second application, data indicating the particular suggested candidate input.

Existing music recognition applications require a connection to a server that performs the actual recognition. In this paper we present a low-power music recognizer that runs entirely on a mobile device and automatically recognizes music without user interaction. To reduce battery consumption, a small music detector runs continuously on the mobile device's DSP chip and wakes up the main application processor only when it is confident that music is present. Once woken, the recognizer on the application processor is provided with a few seconds of audio which is fingerprinted and compared to the stored fingerprints in the on-device fingerprint database of tens of thousands of songs. Our presented system, Now Playing, has a daily battery usage of less than 1% on average, respects user privacy by running entirely on-device and can passively recognize a wide range of music.

Live search for mobile is a cellphone application that allows users to interact with Web-based information portals. Currently the implementation is focused on information related to local businesses: their phone numbers and addresses, directions, reviews, maps of the surrounding area, and traffic. This paper describes a speech-recognition interface which was recently developed for the application, which allows the users to interact by voice. The paper presents the overall architecture, the user interface, the design and implementation of the speech recognition grammars, and initial performance results indicating that for sentence level utterance recognition we achieve 60 to 65% of human capability.

In this paper we address two related challenges in multimodal local search applications on mobile devices: first, correctly displaying the business names, and second, harvesting language model training data from an inconsistently labeled corpus. We investigate the impact of common text normalization and the quality of language model training corpus on the accuracy of displayed results. We propose a new language model framework that eliminates the need for explicit inverse text normalization. The same framework can be applied to sift through corrupted language model training data. Our new language model is 25% more accurate while 25% smaller in size.