Methods for Low Bitrate Coding Enhancement Part I ......Loudness normalization (ITU‐R BS.1770) Three different codec implementations 1.MP3 encoder, 64 kbps, bandwidth 10 kHz (less
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Perceptual audio coders such as MP3 and AAC are used for efficient Storage Transmission
With good encoder and sufficiently high bitrates Uncoded and coded signals are perceptually indistinguishable!
Perceptual transparency At very low bitrates, sound quality is degraded Today`s car audio systems: Lots of possible audio sources with varying audio quality, e.g. Streaming, satellite radio Personal audio collection, etc.
High quality car audio systems unsparingly reveal coding artifacts
Post‐processing of decoded signal without using meta data Easy, versatile integration Can handle transcoding Repeated encoding of already encoded signals E.g. MP3 at 32 kbps is decoded and then encoded with
AAC at 256 kbps Only meta data of the last encoder will be shown!
Enhancement algorithms shall increase perceptual pleasantness In contrast to perceptual transparency
Remove of perceptual audio coding artifacts at low bitrates
3. Bandwidth ExtensionCopy Up and Spectral Envelope Shaping
Copy up of complex spectral coefficients Source patch: [fc, fc‐w] Destination patches: [fc+(k‐1)w, fc+kw] with k = 1,2,3,…
In order to avoid discontinuities between patches Attenuate destination patches by factor T Factor T is based on spectral tilt computed on source patch
Aim: suppress roughness which is caused by copy up This is achieved by filtering of destination patches Two nested allpass filters in parallel with short impulse responses Mainly beneficial for sustained signal part. Therefore the transient‐sustained
decomposition is required
PG3
Slide 11
PG3 Bild von Allpässen?Impulsantwort?Patrick Gampp, 9/5/2017
Can occur at low bitrates Groups of spectral bins are very coarsely quantized to Very low spectral magnitudes (spectral gap) Very high spectral magnitudes (spectral island)
Over time spectral gaps and islands may toggle or can occur isolated Introduction of temporal modulation Artifact is reminiscent of a bird`s twitter
Separate processing of Spectral islands Spectral gaps
Performed in quiet listening room using closed studio headphones 11 listeners Listening test with multiple stimuli Seven‐grade comparison test scale (ITU‐R BS.1284)
Results per item, for all items and groups of codecs Proposed algorithms improve sound quality over all items Speech item (Spe3) also benefits of enhancement
Methods for Restoration of spectral islands and gaps Bandwidth extension
Methods do not use meta data Methods work in real‐time Evaluation in a listening test yielded Proposed algorithms improve sound quality over all items Speech also benefits of proposed algorithms