Commits

Branch selector

 dr-phasenn

User selector

All users

Datepicker

All time

Commit History

Commits on Aug 19, 2019

• this version works over a 0 tp pi/4, better than 20dB, using larer batches and more training data, more epochs

  drowe67

  committed
  f0b053e

  Copy full SHA for f0b053e

Commits on Aug 18, 2019

• loss < 0.01 w between pi/8 and pi/4

  drowe67

  committed
  e9aa03e

  Copy full SHA for e9aa03e

• loss 0.0265 w between 0 and pi/4

  drowe67

  committed
  a62f8c3

  Copy full SHA for a62f8c3

Commits on Aug 16, 2019

• first pass at test1 - doing a reasonable job at predicting outputr phases when quantised to 16 initial phases and 16 frequencies

  drowe67

  committed
  d4e9fbc

  Copy full SHA for d4e9fbc

• test NN for working with phases

  drowe67

  committed
  4da7235

  Copy full SHA for 4da7235

• rm some old octave files

  drowe67

  committed
  c6dc1c8

  Copy full SHA for c6dc1c8

Commits on Aug 13, 2019

• fixed mapping to sparse harmonics

  drowe67

  committed
  936c2ff

  Copy full SHA for 936c2ff

Commits on Aug 10, 2019

• building up phase NN training model

  drowe67

  committed
  8d5fe65

  Copy full SHA for 8d5fe65

• max_amp calculation incorrect

  drowe67

  committed
  da7fb75

  Copy full SHA for da7fb75

Commits on Jul 25, 2019

• Merge pull request #57 from drowe67/ss-fs-offset

  Show description for b1ff0a5
  drowe67

  authored
  b1ff0a5

  Copy full SHA for b1ff0a5

• Didn't mean to remove assert

  

  srsampson

  committed
  70f3b81

  Copy full SHA for 70f3b81

• Minor changes to remove double promotion warnings

  

  srsampson

  committed
  a960248

  Copy full SHA for a960248

Commits on Jul 23, 2019

• Merge pull request #55 from drowe67/dr-vqeq

  Show description for 3568991
  drowe67

  authored
  3568991

  Copy full SHA for 3568991

• updated fig for blog post

  drowe67

  committed
  66fe723

  Copy full SHA for 66fe723

Commits on Jul 21, 2019

• VQEQ working in C code, C output samples sound very similar to Octave simulation, doing a reasonable job at making speech sound better

  drowe67

  committed
  de4eab1

  Copy full SHA for de4eab1

• first pass at equaliser in C

  drowe67

  committed
  3cbc01c

  Copy full SHA for 3cbc01c

• trying a non function script

  drowe67

  committed
  637bf52

  Copy full SHA for 637bf52

• adding octave version of EQ

  drowe67

  committed
  3147720

  Copy full SHA for 3147720

• debugging ctest on Travis

  drowe67

  committed
  4a1108c

  Copy full SHA for 4a1108c

• made test less verbose

  drowe67

  committed
  3317031

  Copy full SHA for 3317031

• cleaning up some unused code

  drowe67

  committed
  8e15967

  Copy full SHA for 8e15967

• debugging ctest

  drowe67

  committed
  80e0b55

  Copy full SHA for 80e0b55

• debugging ctest

  drowe67

  committed
  00e6153

  Copy full SHA for 00e6153

• new ctest for codec 2 700C Octave-C port

  drowe67

  committed
  42560c9

  Copy full SHA for 42560c9

• adaptive front EQ, added wieght for train_120 VQ 20th sample, which neatly brought down obj measures. Listening tests meets goals - samples with LF energy are improved

  drowe67

  committed
  2de9762

  Copy full SHA for 2de9762

Commits on Jul 20, 2019

• better results with a different ideal profile that penaliss high LF energy. improves cq_freedv_8k samples, cq_ref, vk5qi, and doesn't affect other samples.

  drowe67

  committed
  8b7a08d

  Copy full SHA for 8b7a08d

• integrator with gain, but still unsatisfactory on ve9qrp or speed, trying input EQ again

  drowe67

  committed
  1154193

  Copy full SHA for 1154193

• block adaptive working pretty well, except one sample (ve9qrp_10s) messed up by EQ

  drowe67

  committed
  4403042

  Copy full SHA for 4403042

• iterative version sounds pretty good and gd obj measures

  drowe67

  committed
  b51a4b9

  Copy full SHA for b51a4b9

Commits on Jul 19, 2019

• experimenting with block based technqiues

  drowe67

  committed
  93d38b0

  Copy full SHA for 93d38b0

• moved EQ in front of VQ, based on differnece between mean of targets vectors and stage 1 of VQ. Good obj results using block averages

  drowe67

  committed
  27ed608

  Copy full SHA for 27ed608

• adaptive eq experients, including contrived lfboost/hfcut experiments. lfboost sound much worse, and EQ has only a small effect

  drowe67

  committed
  6b29818

  Copy full SHA for 6b29818

Commits on Jul 18, 2019

• building up Octave simulation of adaptive EQ

  drowe67

  committed
  053a7bc

  Copy full SHA for 053a7bc

• Merge pull request #53 from drowe67/dr-sm1000

  Show description for 9c4d23d
  drowe67

  authored
  9c4d23d

  Copy full SHA for 9c4d23d

Commits on Jul 17, 2019

• minor tweaks

  drowe67

  committed
  195a989

  Copy full SHA for 195a989

Pagination

Previous

Next