Skip to content

Conversation

@drowe67
Copy link
Owner

@drowe67 drowe67 commented Nov 5, 2021

This PR supports R&D aimed at improving Codec 2 speech quality (WP2000 in ARDC plan). It's not intended for merge, rather to support R&D. The development is conducted using a combination of C, Octave and bash scripts. The R&D is documented in a LaTeX report in codec2/doc/ratek_resampler.

Topics include:

  1. ratek resampling.
  2. The effect of PAPR on speech quality.
  3. The drop in speech quality when coarsely quantised amplitudes are used to synthesise phase spectra.
  4. Two new post filters.
  5. Vector quantisation of rate K vectors.
  6. Microphone equalisation.
  7. Pitch errors and background noise problems
  8. Energy Normalisation.
  9. Mean versus energy quantisation.
  10. SSB style compression of Hilbert Clipping of synthesised speech.
  11. Candidates for new 700 and 1200 bits/s modes.
  12. High Pass filtering of input speech.

@drowe67 drowe67 changed the title Codec 2 PAPR experiments Codec 2 PAPR and Phase experiments Apr 25, 2022
drowe67 added 30 commits March 24, 2023 10:28
…mples, however messes up background noise handling in mmt1 as raised energy is above threshold in postfilter.c
…ression using hilbert clipper on synthesis. Sounds pretty good (even mmt1) with 10dB gain
…Seems to work OK by HC is complicating comparison
…ver now quantised mean samples are softer, and different segments have different levels. Not satisfactory
… on 27 and 12 bit VQ, although level is a little lower. This is a form of compressions. Level comes up nicely with 10dB synthesis/HC gain
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants