A Time-Warped MDCT Approach to Speech Transform Coding
The modified discrete cosine transform (MDCT) is often used for audio coding due to its critical sampling property and good energy compaction, especially for harmonic tones with constant fundamental frequencies (pitch). However, in voiced human speech the pitch is time-varying and thus the energy is spread over several transform coefficients, leading to a reduction of coding efficiency. The approach presented herein compensates for pitch variation in each MDCT block by application of time-variant re-sampling. A dedicated signal adaptive transform window computation ensures the preservation of the time domain aliasing cancelation (TDAC) property. Re-sampling can be designed such that the duration of the processed blocks is not altered, facilitating the replacement of the conventional MDCT in existing audio coders.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is temporarily free for AES members.