Acoustical and Environmental Robustness in Automatic Speech by Alejandro Acero (auth.)

By Alejandro Acero (auth.)

The desire for automated speech reputation platforms to be powerful with appreciate to adjustments of their acoustical atmosphere has develop into extra largely liked in recent times, as extra structures are discovering their manner into sensible functions. even if the difficulty of environmental robustness has got just a small fraction of the eye dedicated to speaker independence, even speech reputation platforms which are designed to be speaker self sustaining usually practice very poorly after they are verified utilizing a special kind of microphone or acoustical setting from the only with which they have been informed. using microphones except a "close­ speaking" headset additionally has a tendency to critically degrade speech popularity -performance. Even in rather quiet place of work environments, speech is degraded via additive noise from enthusiasts, slamming doorways, and different conversations, in addition to by way of the results of unknown linear filtering bobbing up reverberation from floor reflections in a room, or spectral shaping by means of microphones or the vocal tracts of person audio system. Speech-recognition platforms designed for long-distance mobile traces, or functions deployed in additional hostile acoustical environments comparable to motorcars, manufacturing facility flooring, oroutdoors call for some distance greaterdegrees ofenvironmental robustness. There are numerous alternative ways of creating acoustical robustness into speech reputation structures. Arrays of microphones can be utilized to enhance a directionally-sensitive approach that resists intelference from competing talkers and different noise assets which are spatially separated from the resource of the specified speech signal.

Show description

Read Online or Download Acoustical and Environmental Robustness in Automatic Speech Recognition PDF

Best acoustics & sound books

Noise reduction in speech processing

Noise is far and wide and in such a lot functions which are with regards to audio and speech, corresponding to human-machine interfaces, hands-free communications, voice over IP (VoIP), listening to aids, teleconferencing/telepresence/telecollaboration structures, and such a lot of others, the sign of curiosity (usually speech) that's picked up via a microphone is mostly infected via noise.

Understanding and Crafting the Mix, : The Art of Recording

Realizing and Crafting the combo delivers transparent and systematic equipment for opting for, comparing, and shaping the creative parts in tune and audio recording. The workouts all through assist you to boost severe listening and comparing talents and achieve better keep watch over over the standard of your recordings.

Noise Control From Concept to Application

A good studying instrument for college kids and practitioners, this advisor to noise keep an eye on will permit readers to take advantage of their wisdom to unravel a variety of business noise keep watch over difficulties. operating from simple medical rules, the writer exhibits how an realizing of sound may be utilized to real-world settings, operating via numerous examples intimately and masking strong perform in noise keep an eye on for either new and current amenities.

Additional resources for Acoustical and Environmental Robustness in Automatic Speech Recognition

Sample text

5. The Recognition System We also performed these evaluations using a more compact and easilytrained version of Sphinx with only 329 triphone models, omitting such features as duration, function-word and function-phrase models, betweenword triphone models, and corrective training. We were willing to tolerate the somewhat lower absolute recognition accuracy that this version of Sphinx provided because of the reduced time required by the training process. Using the census database, the more compact Sphinx system, and DEC 3100 tSperplexiry is an information theoretic measure of the amount of constraint imposed by a finite-state grammar.

4]), there has been a great deal of recent interest in its application to robust speech recognition. In the latter case, the end user of the processed speech is not a human being but a computer. We present in this section an introduction to spectral subtraction with its use in speech enhancement and recognition. Then a framework for processing in the log-spectrum is presented. 1. Spectral Subtraction for Speech Enhancement Spectral subtraction is a family of techniques that attempt to subtract the noise energy from the noisy speech energy at every frequency band.

To this end, we recorded a training database stereophonically using two different microphones: a close-talking microphone and a desk-top microphone. Another common characterization of noisy speech databases is that of SNR. Although under some circumstances SNR can provide a good estimate of the degree of difficulty of a speech database that has been recorded in the presence of white noise, it does not characterize the database when the noise is colored or when the speech has been passed through a filter that has altered the frequency response, as is the case for real environments.

Download PDF sample

Rated 4.18 of 5 – based on 37 votes

Categories: Acoustics Sound