Showing posts with label Speech. Show all posts
Showing posts with label Speech. Show all posts

Monday, 16 April 2007

Audio Watermarking Algorithm for Copyright Protection

(Photo quoted from NeAsia)

Digital watermark technology is now drawing attention as a new method of protecting digital content from unauthorized one.

Critical band
It can be shown that the sensitivity of the ear to every frequency is not quite the same.



Masking effect
Frequency domain masking also called simultaneous masking. If two signals have close frequency, the louder one will make the weaker one inaudible. For example, conversation at a bus stop can be completely impossible if a loud bus is driving past. The masking phenomenon occurs because any loud sound will distort the Absolute Threshold of Hearing, making quieter, otherwise perceptible sounds inaudible.

DSSS
(Wikipedia)In telecommunications, direct-sequence spread spectrum (DSSS) is a modulation technique. As with other spread-spectrum technologies, the transmitted signal takes up more bandwidth than the information signal that is being modulated. The name 'spread spectrum' comes from the fact that the carrier signals occur over the full bandwidth (spectrum) of a device's transmitting frequency.
Math formula Tb = NTc, where N is the length of PN sequence

Watermark embedding
(Quoted from MusicTrace) With the aid of the software products of the ContentMark product family, it is possible to embed additional information into audio signals. This additional information is transmitted to the final user hidden in the music in a form that is imperceptible to human hearing. A further characteristic is the fact that embedding of a watermark does not change the format. The final user does not have to purchase special player devices, instead he can still play these titles using conventional equipment.
The additional data are robustly embedded in the audio signal; this means that they cannot be removed by simple means. The objective of developing the audio watermark technique was to ensure that the watermark does not become unusable until intentional or inadvertent disturbances have degraded the audio quality to such an extent that the recorded title no longer has any economic value.
The information to be embedded is transferred in so-called data containers. Several data containers have already been developed, they differ in the volume of data that is to be transferred, the data rate and the robustness of the watermark. The two most commonly used data containers transfer 48 bits in 5 or 2.7 seconds. Other data containers can be generated as agreed with the customer.

Watermark extraction
(quoted from http://www.leeds.ac.uk/)Paper-based Watermark Extraction project (WatEx) was established in October 2004 as a Ph.D. research study, the aim of this project is to automatically retrieve and store paper watermarks in a digital representation in order to preserve its historical value, and to provide better access and distribution with the current Information and Computing Technology (ICT). The focus will be on the digital acquisition, and automatic processing and analysis of the visible paper-based watermark, probing beyond the paper surface data to extract the watermark design and to create a digital representation for long term preservation.



Sunday, 1 April 2007

Spontaneous Speech

(photo quoted from Nippon Telegraph and Telephone Corporation )
Introduction
Spontaneous speech requires no special training, efficient, minimal cognitive load and wealth of information at multiple levels. There are some features such as variable speaking rate, greater use of short words and function words, and a tendency towards more coarticulation and generally less precise coarticulation.

Challenge
1. Recovering hidden punctuation
Punctuation is everything in written language other than the actual letters, including punctuation marks and inter-word spaces.
2. Coping with disfluences
Disfluency: Lack of skilfulness in speaking or writing, such as filled pauses, repetitions, repairs, false starts
3. Allowing for realistic turn-taking
Listeners project the end of a current speaker’s turn using syntax, semantics, pragmatics, and prosody.
4. Hearing more than words
Distinguish emotion and do user state detection. Especially, it’s important for certain dialog system application