issues
search
vvolhejn
/
thesis
ETH Zürich MSc Thesis: Accelerating Neural Audio Synthesis
Apache License 2.0
17
stars
1
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Re-run experiments with quantized models after new setting
#49
vvolhejn
closed
2 years ago
1
Debug quantization
#48
vvolhejn
closed
2 years ago
1
Train on URMP trumpet dataset
#47
vvolhejn
closed
2 years ago
0
Find out what's wrong with Ault
#46
vvolhejn
closed
2 years ago
1
Debug NEWT
#45
vvolhejn
closed
2 years ago
0
Re-run dilated conv benchmarks with deepsparse-nightly
#44
vvolhejn
closed
2 years ago
1
Benchmark dilated conv with two stacks
#43
vvolhejn
closed
2 years ago
1
Benchmark with different number of CPUs/vCPUs
#42
vvolhejn
opened
2 years ago
0
Look into TVM+sparsity
#41
vvolhejn
closed
2 years ago
2
Try block quantization for TFLite
#40
vvolhejn
closed
2 years ago
1
Run optimized model as part of DDSP and RAVE
#39
vvolhejn
closed
2 years ago
1
Conduct mean opinion score survey
#38
vvolhejn
opened
2 years ago
1
Try depthwise-separable dilated convolution
#37
vvolhejn
closed
2 years ago
2
Measure the overhead of Python vs C++
#36
vvolhejn
closed
2 years ago
1
Try on Tensorflow Metal using Apple's GPU
#35
vvolhejn
opened
2 years ago
0
Increase size of dense models to get full CPU utilization
#34
vvolhejn
closed
2 years ago
2
Try tuning for TVM
#33
vvolhejn
closed
2 years ago
3
Try pruning for DeepSparse
#32
vvolhejn
closed
2 years ago
0
Add quantization to DeepSparse
#31
vvolhejn
closed
2 years ago
0
Add quantization to TVM
#30
vvolhejn
closed
2 years ago
0
Create PyTorch version of dilated convolution block
#29
vvolhejn
closed
2 years ago
0
Make convolutions in DilatedConvDecoder causal
#28
vvolhejn
closed
2 years ago
1
Compute losses in a way that's not included in the timing
#27
vvolhejn
closed
2 years ago
0
Use multiband decomposition in GAN discriminator
#26
vvolhejn
opened
2 years ago
0
Include CREPE encoding time in evaluation
#25
vvolhejn
closed
2 years ago
1
Pass f0 confidence to DDSP-based models
#24
vvolhejn
closed
2 years ago
0
Log timbre transfer results
#23
vvolhejn
closed
2 years ago
0
Make W&B calls have the proper step number
#22
vvolhejn
closed
2 years ago
0
Add adversarial loss
#21
vvolhejn
opened
2 years ago
0
Implement FastNEWT
#20
vvolhejn
closed
2 years ago
0
Run model once before creating W&B run
#19
vvolhejn
closed
2 years ago
0
Name Wandb run after what prepare_job creates
#18
vvolhejn
closed
2 years ago
1
Add a test set
#17
vvolhejn
closed
2 years ago
1
Add data augmentation
#16
vvolhejn
opened
2 years ago
1
Implement RAVE
#15
vvolhejn
closed
2 years ago
1
Use Weights and Biases or similar?
#14
vvolhejn
closed
2 years ago
1
Implement multiband decomposition
#13
vvolhejn
closed
2 years ago
1
Look into datasets
#11
vvolhejn
opened
2 years ago
4
Try other architectural modifications (like in NEWT)
#9
vvolhejn
closed
2 years ago
1
Try replacing the harmonics-plus-noise model altogether
#8
vvolhejn
closed
2 years ago
1
Try non-autoregressive options instead of GRU
#7
vvolhejn
closed
2 years ago
0
Benchmark/profile baselines to identify performance bottlenecks
#6
vvolhejn
closed
2 years ago
1
Automated evaluation using e.g. Fréchet Audio Distance
#5
vvolhejn
closed
2 years ago
1
Rewrite NEWT using DDSP+TensorFlow
#4
vvolhejn
closed
2 years ago
2
Setup DDSP code
#3
vvolhejn
closed
2 years ago
1
Choose a framework - Pytorch vs TensorFlow
#2
vvolhejn
closed
2 years ago
1
Choose a license
#1
vvolhejn
closed
2 years ago
0