diff options
| -rw-r--r-- | .gitignore | 8 | ||||
| -rw-r--r-- | best-epoch.sh | 1 | ||||
| -rwxr-xr-x | bin/crossfade_cat.sh | 129 | ||||
| -rwxr-xr-x | bin/loop.sh | 20 | ||||
| -rwxr-xr-x | bin/xfade.sh | 23 | ||||
| -rwxr-xr-x | datasets/count_subdirs.sh | 1 | ||||
| -rwxr-xr-x | datasets/generate.sh | 34 | ||||
| -rwxr-xr-x | datasets/split44k.sh | 24 | ||||
| -rwxr-xr-x | datasets/spread.sh | 25 | ||||
| -rw-r--r-- | generate.py | 365 | ||||
| -rw-r--r-- | glass_test2.sh | 85 | ||||
| -rw-r--r-- | glass_test3.sh | 112 | ||||
| -rw-r--r-- | glass_tests.sh | 202 | ||||
| -rw-r--r-- | kick_test.sh | 51 | ||||
| -rwxr-xr-x | mix.sh | 20 | ||||
| -rw-r--r-- | train.py | 2 | ||||
| -rw-r--r-- | train_basic_22k.sh | 15 | ||||
| -rw-r--r-- | train_basic_32k.sh | 15 | ||||
| -rw-r--r-- | train_basic_44k.sh | 15 | ||||
| -rw-r--r-- | train_drums.sh | 44 | ||||
| -rw-r--r-- | train_test_generate.sh | 93 | ||||
| -rw-r--r-- | trainer/__init__.py | 9 | ||||
| -rw-r--r-- | trainer/plugins.py | 8 |
23 files changed, 1296 insertions, 5 deletions
@@ -92,3 +92,11 @@ ENV/ *~ *.swp *.swo + +*.wav +*.mp3 +*.aif +*.aiff + +results/ + diff --git a/best-epoch.sh b/best-epoch.sh new file mode 100644 index 0000000..230daa0 --- /dev/null +++ b/best-epoch.sh @@ -0,0 +1 @@ +ls -lat results/*/checkpoints/best* | head -n 20 diff --git a/bin/crossfade_cat.sh b/bin/crossfade_cat.sh new file mode 100755 index 0000000..13a8e90 --- /dev/null +++ b/bin/crossfade_cat.sh @@ -0,0 +1,129 @@ +#!/bin/bash +# +# crossfade_cat.sh +# +# Concatenates two files together with a crossfade of $1 seconds. +# Filenames are specified as $2 and $3. +# +# $4 is optional and specifies if a fadeout should be performed on +# first file. +# $5 is optional and specifies if a fadein should be performed on +# second file. +# +# Example: crossfade_cat.sh 10 infile1.wav infile2.wav auto auto +# +# By default, the script attempts to guess if the audio files +# already have a fadein/out on them or if they just have really +# low volumes that won't cause clipping when mixxing. If this +# is not detected then the script will perform a fade in/out to +# prevent clipping. +# +# The user may specify "yes" or "no" to force the fade in/out +# to occur. They can also specify "auto" which is the default. +# +# Crossfaded file is created as "mix.wav". +# +# Original script from Kester Clegg. Mods by Chris Bagwell to show +# more examples of sox features. +# + +SOX=sox +SOXI=soxi + +if [ "$3" == "" ]; then + echo "Usage: $0 crossfade_seconds first_file second_file [ fadeout ] [ fadein ]" + echo + echo "If a fadeout or fadein is not desired then specify \"no\" for that option. \"yes\" will force a fade and \"auto\" will try to detect if a fade should occur." + echo + echo "Example: $0 10 infile1.wav infile2.wav auto auto" + exit 1 +fi + +fade_length=$1 +first_file=$2 +second_file=$3 + +fade_first="auto" +if [ "$4" != "" ]; then + fade_first=$4 +fi + +fade_second="auto" +if [ "$5" != "" ]; then + fade_second=$5 +fi + +fade_first_opts= +if [ "$fade_first" != "no" ]; then + fade_first_opts="fade t 0 0:0:$fade_length 0:0:$fade_length" +fi + +fade_second_opts= +if [ "$fade_second" != "no" ]; then + fade_second_opts="fade t 0:0:$fade_length" +fi + +echo "crossfade and concatenate files" +echo +echo "Finding length of $first_file..." +first_length=`$SOX "$first_file" 2>&1 -n stat | grep Length | cut -d : -f 2 | cut -f 1` +echo "Length is $first_length seconds" + +trim_length=`echo "$first_length - $fade_length" | bc` + +# Get crossfade section from first file and optionally do the fade out +echo "Obtaining $fade_length seconds of fade out portion from $first_file..." +$SOX "$first_file" -e signed-integer -b 16 fadeout1.wav trim $trim_length + +# When user specifies "auto" try to guess if a fadeout is needed. +# "RMS amplitude" from the stat effect is effectively an average +# value of samples for the whole fade length file. If it seems +# quite then assume a fadeout has already been done. An RMS value +# of 0.1 was just obtained from trail and error. +if [ "$fade_first" == "auto" ]; then + RMS=`$SOX fadeout1.wav 2>&1 -n stat | grep RMS | grep amplitude | cut -d : -f 2 | cut -f 1` + should_fade=`echo "$RMS > 0.1" | bc` + if [ $should_fade == 0 ]; then + echo "Auto mode decided not to fadeout with RMS of $RMS" + fade_first_opts="" + else + echo "Auto mode will fadeout" + fi +fi + +$SOX fadeout1.wav fadeout2.wav $fade_first_opts + +# Get the crossfade section from the second file and optionally do the fade in +echo "Obtaining $fade_length seconds of fade in portion from $second_file..." +$SOX "$second_file" -e signed-integer -b 16 fadein1.wav trim 0 $fade_length + +# For auto, do similar thing as for fadeout. +if [ "$fade_second" == "auto" ]; then + RMS=`$SOX fadein1.wav 2>&1 -n stat | grep RMS | grep amplitude | cut -d : -f 2 | cut -f 1` + should_fade=`echo "$RMS > 0.1" | bc` + if [ $should_fade == 0 ]; then + echo "Auto mode decided not to fadein with RMS of $RMS" + fade_second_opts="" + else + echo "Auto mode will fadein" + fi +fi + +$SOX fadein1.wav fadein2.wav $fade_second_opts + +# Mix the crossfade files together at full volume +echo "Crossfading..." +$SOX -m -v 1.0 fadeout2.wav -v 1.0 fadein2.wav crossfade.wav + +echo "Trimming off crossfade sections from original files..." + +$SOX "$first_file" -e signed-integer -b 16 song1.wav trim 0 $trim_length +$SOX "$second_file" -e signed-integer -b 16 song2.wav trim $fade_length +$SOX song1.wav crossfade.wav song2.wav mix.wav + +echo -e "Removing temporary files...\n" +rm fadeout1.wav fadeout2.wav fadein1.wav fadein2.wav crossfade.wav song1.wav song2.wav +mins=`echo "$trim_length / 60" | bc` +secs=`echo "$trim_length % 60" | bc` +echo "The crossfade in mix.wav occurs at around $mins mins $secs secs" + diff --git a/bin/loop.sh b/bin/loop.sh new file mode 100755 index 0000000..beb1286 --- /dev/null +++ b/bin/loop.sh @@ -0,0 +1,20 @@ +rm fades.txt +rm files.txt +rm mix.sh + +ITER=0 +NEXT=1 +for i in `ls -v *.wav` +do +echo "[0$ITER][$NEXT]acrossfade=ns=5:o=1:c1=tri:c2=tri[0$NEXT];" >> fades.txt +printf '\-i %s ' $i >> files.txt +ITER=$(expr $ITER + 1) +NEXT=$(expr $NEXT + 1) +done + +printf "ffmpeg " >> mix.sh +cat files.txt >> mix.sh +printf " -filter_complex \"" >> mix.sh +cat fades.txt >> mix.sh +echo '\" out.wav' >> mix.sh + diff --git a/bin/xfade.sh b/bin/xfade.sh new file mode 100755 index 0000000..5e861e0 --- /dev/null +++ b/bin/xfade.sh @@ -0,0 +1,23 @@ +crossfade_dur=1 +i=0 +limit=10000 + +for file in `ls *.wav | sort -V` +do + i=$((i+1)) + if [ $i -eq $limit ] + then + break + fi + + if [ $i -eq 1 ] + then + cp $file mix.wav + else + # ../../crossfade_cat.sh $crossfade_dur mix.wav $file yes yes + echo $file + sox mix.wav "$file" out.wav splice $(soxi -D mix.wav),0.01 + mv out.wav mix.wav + fi +done + diff --git a/datasets/count_subdirs.sh b/datasets/count_subdirs.sh new file mode 100755 index 0000000..3999b3c --- /dev/null +++ b/datasets/count_subdirs.sh @@ -0,0 +1 @@ +find -maxdepth 1 -type d | sort | while read -r dir; do printf "%s:\t" "$dir"; find "$dir" -type f | wc -l; done diff --git a/datasets/generate.sh b/datasets/generate.sh new file mode 100755 index 0000000..335928c --- /dev/null +++ b/datasets/generate.sh @@ -0,0 +1,34 @@ +function process () { + echo "____________________________________________________" + echo "process $1" + name=$1 + in="${name}.wav" + out="s_${in}" + ./spread.sh $in $out 0.99 0.01 1.01 + ./split44k.sh $out 8 "44k_$name" + rm $out +} +function ease_process () { + echo "____________________________________________________" + echo "ease_process $1" + name=$1 + step=$2 + in="${name}.wav" + sout="o_${in}" + out="s_${in}" + sox -v 0.95 $in $sout + ./spread.sh $sout $out 0.999 $step 1.001 + ./split44k.sh $out 8 "44k_$name" + rm $sout + rm $out +} +#ease_process '' 0.0000 +ease_process 'blblbl' 0.00001515 +ease_process 'faty-scrub1' 0.0000285 +ease_process 'faty-medieval' 0.00003 +ease_process 'faty-crystals' 0.0000111 +ease_process 'faty-vocal1' 0.000013 +ease_process 'faty-vocal2' 0.000028145 +ease_process 'faty-scrub2' 0.00000466 +ease_process 'siren' 0.0000275 + diff --git a/datasets/split44k.sh b/datasets/split44k.sh new file mode 100755 index 0000000..4884af1 --- /dev/null +++ b/datasets/split44k.sh @@ -0,0 +1,24 @@ +#/bin/sh + +if [ "$#" -ne 3 ]; then + echo "Usage: $0 <filename.wav> <chunk size in seconds> <dataset path>" + exit +fi + +fn=$1 +chunk_size=$2 +dataset_path=$3 + +converted=".temp2.wav" +rm -f $converted +ffmpeg -i $fn -ac 1 -ar 44100 $converted + +mkdir $dataset_path +length=$(ffprobe -i $converted -show_entries format=duration -v quiet -of csv="p=0") +end=$(echo "$length / $chunk_size - 1" | bc) +echo "splitting..." +for i in $(seq 0 $end); do + ffmpeg -hide_banner -loglevel error -ss $(($i * $chunk_size)) -t $chunk_size -i $converted "$dataset_path/$i.wav" +done +echo "done" +rm -f $converted diff --git a/datasets/spread.sh b/datasets/spread.sh new file mode 100755 index 0000000..bec1da3 --- /dev/null +++ b/datasets/spread.sh @@ -0,0 +1,25 @@ + +if [ "$#" -ne 5 ]; then + echo "Usage: $0 <in.wav> <out.wav> <rate_min> <rate_step> <rate_max>" + exit +fi + +FN_IN=$1 +FN_OUT=$2 +RATE=$3 +STEP=$4 +MAX=$5 + +ITER=0 +while true; do + if (( $(echo "$RATE > $MAX" | bc -l) )); then + break + fi + let ITER+=1 + RATE=`echo "$RATE+$STEP" | bc` + sox $FN_IN "tmp_$ITER.wav" speed $RATE +done + +sox tmp_* $FN_OUT +rm tmp_* + diff --git a/generate.py b/generate.py new file mode 100644 index 0000000..92a930f --- /dev/null +++ b/generate.py @@ -0,0 +1,365 @@ +# CometML needs to be imported first. +try: + import comet_ml +except ImportError: + pass + +from model import SampleRNN, Predictor +from optim import gradient_clipping +from nn import sequence_nll_loss_bits +from trainer import Trainer +from trainer.plugins import ( + TrainingLossMonitor, ValidationPlugin, AbsoluteTimeMonitor, SaverPlugin, + GeneratorPlugin, StatsPlugin +) +from dataset import FolderDataset, DataLoader + +import torch +from torch.utils.trainer.plugins import Logger + +from natsort import natsorted + +from functools import reduce +import os +import shutil +import sys +from glob import glob +import re +import argparse + + +default_params = { + # model parameters + 'n_rnn': 1, + 'dim': 1024, + 'learn_h0': True, + 'q_levels': 256, + 'seq_len': 1024, + 'weight_norm': True, + 'batch_size': 128, + 'val_frac': 0.1, + 'test_frac': 0.1, + + # training parameters + 'keep_old_checkpoints': False, + 'datasets_path': 'datasets', + 'results_path': 'results', + 'epoch_limit': 1000, + 'resume': True, + 'sample_rate': 16000, + 'n_samples': 1, + 'sample_length': 80000, + 'loss_smoothing': 0.99, + 'cuda': True, + 'comet_key': None +} + +tag_params = [ + 'exp', 'frame_sizes', 'n_rnn', 'dim', 'learn_h0', 'q_levels', 'seq_len', + 'batch_size', 'dataset', 'val_frac', 'test_frac' +] + +def param_to_string(value): + if isinstance(value, bool): + return 'T' if value else 'F' + elif isinstance(value, list): + return ','.join(map(param_to_string, value)) + else: + return str(value) + +def make_tag(params): + return '-'.join( + key + ':' + param_to_string(params[key]) + for key in tag_params + if key not in default_params or params[key] != default_params[key] + ) + +def setup_results_dir(params): + def ensure_dir_exists(path): + if not os.path.exists(path): + os.makedirs(path) + + tag = make_tag(params) + results_path = os.path.abspath(params['results_path']) + ensure_dir_exists(results_path) + results_path = os.path.join(results_path, tag) + if not os.path.exists(results_path): + os.makedirs(results_path) + elif not params['resume']: + shutil.rmtree(results_path) + os.makedirs(results_path) + + for subdir in ['checkpoints', 'samples']: + ensure_dir_exists(os.path.join(results_path, subdir)) + + return results_path + +def load_last_checkpoint(checkpoints_path): + checkpoints_pattern = os.path.join( + checkpoints_path, SaverPlugin.last_pattern.format('*', '*') + ) + checkpoint_paths = natsorted(glob(checkpoints_pattern)) + if len(checkpoint_paths) > 0: + checkpoint_path = checkpoint_paths[-1] + checkpoint_name = os.path.basename(checkpoint_path) + match = re.match( + SaverPlugin.last_pattern.format(r'(\d+)', r'(\d+)'), + checkpoint_name + ) + epoch = int(match.group(1)) + iteration = int(match.group(2)) + return (torch.load(checkpoint_path), epoch, iteration) + else: + return None + +def tee_stdout(log_path): + log_file = open(log_path, 'a', 1) + stdout = sys.stdout + + class Tee: + + def write(self, string): + log_file.write(string) + stdout.write(string) + + def flush(self): + log_file.flush() + stdout.flush() + + sys.stdout = Tee() + +def make_data_loader(overlap_len, params): + path = os.path.join(params['datasets_path'], params['dataset']) + def data_loader(split_from, split_to, eval): + dataset = FolderDataset( + path, overlap_len, params['q_levels'], split_from, split_to + ) + return DataLoader( + dataset, + batch_size=params['batch_size'], + seq_len=params['seq_len'], + overlap_len=overlap_len, + shuffle=(not eval), + drop_last=(not eval) + ) + return data_loader + +def init_comet(params, trainer): + if params['comet_key'] is not None: + from comet_ml import Experiment + from trainer.plugins import CometPlugin + experiment = Experiment(api_key=params['comet_key'], log_code=False) + hyperparams = { + name: param_to_string(params[name]) for name in tag_params + } + experiment.log_multiple_params(hyperparams) + trainer.register_plugin(CometPlugin( + experiment, [ + ('training_loss', 'epoch_mean'), + 'validation_loss', + 'test_loss' + ] + )) + +def main(exp, frame_sizes, dataset, **params): + params = dict( + default_params, + exp=exp, frame_sizes=frame_sizes, dataset=dataset, + **params + ) + + results_path = setup_results_dir(params) + tee_stdout(os.path.join(results_path, 'log')) + + model = SampleRNN( + frame_sizes=params['frame_sizes'], + n_rnn=params['n_rnn'], + dim=params['dim'], + learn_h0=params['learn_h0'], + q_levels=params['q_levels'], + weight_norm=params['weight_norm'] + ) + predictor = Predictor(model) + if params['cuda']: + model = model.cuda() + predictor = predictor.cuda() + + optimizer = gradient_clipping(torch.optim.Adam(predictor.parameters())) + + data_loader = make_data_loader(model.lookback, params) + test_split = 1 - params['test_frac'] + val_split = test_split - params['val_frac'] + + trainer = Trainer( + predictor, sequence_nll_loss_bits, optimizer, + data_loader(0, val_split, eval=False), + cuda=params['cuda'] + ) + + checkpoints_path = os.path.join(results_path, 'checkpoints') + checkpoint_data = load_last_checkpoint(checkpoints_path) + if checkpoint_data is not None: + (state_dict, epoch, iteration) = checkpoint_data + trainer.epochs = epoch + trainer.iterations = iteration + predictor.load_state_dict(state_dict) + print("epochs: {} iterations: {}".format(epoch, iteration)) + + """ + trainer.register_plugin(TrainingLossMonitor( + smoothing=params['loss_smoothing'] + )) + """ + trainer.register_plugin(ValidationPlugin( + data_loader(val_split, test_split, eval=True), + data_loader(test_split, 1, eval=True) + )) + trainer.register_plugin(AbsoluteTimeMonitor()) + """ + trainer.register_plugin(SaverPlugin( + checkpoints_path, params['keep_old_checkpoints'] + )) + """ + trainer.register_plugin(GeneratorPlugin( + os.path.join(results_path, 'samples'), params['n_samples'], + params['sample_length'], params['sample_rate'] + )) + """ + trainer.register_plugin( + Logger([ + 'training_loss', + 'validation_loss', + 'test_loss', + 'time' + ]) + ) + trainer.register_plugin(StatsPlugin( + results_path, + iteration_fields=[ + 'training_loss', + ('training_loss', 'running_avg'), + 'time' + ], + epoch_fields=[ + 'validation_loss', + 'test_loss', + 'time' + ], + plots={ + 'loss': { + 'x': 'iteration', + 'ys': [ + 'training_loss', + ('training_loss', 'running_avg'), + 'validation_loss', + 'test_loss', + ], + 'log_y': True + } + } + )) + init_comet(params, trainer) + """ + trainer.generate(int(params['epoch_limit'])) + + +if __name__ == '__main__': + parser = argparse.ArgumentParser( + formatter_class=argparse.ArgumentDefaultsHelpFormatter, + argument_default=argparse.SUPPRESS + ) + + def parse_bool(arg): + arg = arg.lower() + if 'true'.startswith(arg): + return True + elif 'false'.startswith(arg): + return False + else: + raise ValueError() + + parser.add_argument('--exp', required=True, help='experiment name') + parser.add_argument( + '--frame_sizes', nargs='+', type=int, required=True, + help='frame sizes in terms of the number of lower tier frames, \ + starting from the lowest RNN tier' + ) + parser.add_argument( + '--dataset', required=True, + help='dataset name - name of a directory in the datasets path \ + (settable by --datasets_path)' + ) + parser.add_argument( + '--n_rnn', type=int, help='number of RNN layers in each tier' + ) + parser.add_argument( + '--dim', type=int, help='number of neurons in every RNN and MLP layer' + ) + parser.add_argument( + '--learn_h0', type=parse_bool, + help='whether to learn the initial states of RNNs' + ) + parser.add_argument( + '--q_levels', type=int, + help='number of bins in quantization of audio samples' + ) + parser.add_argument( + '--seq_len', type=int, + help='how many samples to include in each truncated BPTT pass' + ) + parser.add_argument( + '--weight_norm', type=parse_bool, + help='whether to use weight normalization' + ) + parser.add_argument('--batch_size', type=int, help='batch size') + parser.add_argument( + '--val_frac', type=float, + help='fraction of data to go into the validation set' + ) + parser.add_argument( + '--test_frac', type=float, + help='fraction of data to go into the test set' + ) + parser.add_argument( + '--keep_old_checkpoints', type=parse_bool, + help='whether to keep checkpoints from past epochs' + ) + parser.add_argument( + '--datasets_path', help='path to the directory containing datasets' + ) + parser.add_argument( + '--results_path', help='path to the directory to save the results to' + ) + parser.add_argument('--epoch_limit', help='how many epochs to run') + parser.add_argument( + '--resume', type=parse_bool, default=True, + help='whether to resume training from the last checkpoint' + ) + parser.add_argument( + '--sample_rate', type=int, + help='sample rate of the training data and generated sound' + ) + parser.add_argument( + '--n_samples', type=int, + help='number of samples to generate in each epoch' + ) + parser.add_argument( + '--sample_length', type=int, + help='length of each generated sample (in samples)' + ) + parser.add_argument( + '--loss_smoothing', type=float, + help='smoothing parameter of the exponential moving average over \ + training loss, used in the log and in the loss plot' + ) + parser.add_argument( + '--cuda', type=parse_bool, + help='whether to use CUDA' + ) + parser.add_argument( + '--comet_key', help='comet.ml API key' + ) + + parser.set_defaults(**default_params) + + main(**vars(parser.parse_args())) diff --git a/glass_test2.sh b/glass_test2.sh new file mode 100644 index 0000000..072c451 --- /dev/null +++ b/glass_test2.sh @@ -0,0 +1,85 @@ +function runq () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + fs1=$4 + fs2=$5 + qs=$6 + + exp_name="space_q$qs" + + echo "__________________________________________" + echo "__________________________________________" + echo "__________________________________________" + echo "" + echo ">> running $exp_name" + echo "" + echo "__________________________________________" + echo "__________________________________________" + echo "__________________________________________" + + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes $fs1 $fs2 \ + --n_rnn 2 --dim 1024 --q_levels 512 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 6 \ + --resume True +} +function generateq () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + fs1=$4 + fs2=$5 + qs=$6 + + exp_name="space_q$qs" + + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo "" + echo ">> generating $exp_name" + echo "" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + + python generate.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes $fs1 $fs2 \ + --n_rnn 2 --dim 1024 --q_levels 512 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 6 \ + --resume True +} + +#runq 44k_glass_space1 44100 5 8 2 32 +runq 44k_glass_space1 44100 5 8 2 1 +runq 44k_glass_space1 44100 5 8 2 2 +runq 44k_glass_space1 44100 5 8 2 4 +runq 44k_glass_space1 44100 5 8 2 8 +runq 44k_glass_space1 44100 5 8 2 16 +runq 44k_glass_space1 44100 5 8 2 32 +#runq 44k_glass_space1 44100 5 8 2 2048 + +#generateq 44k_glass_space1 44100 5 8 2 32 +#generateq 44k_glass_space1 44100 5 8 2 1 +#generateq 44k_glass_space1 44100 5 8 2 2 +#generateq 44k_glass_space1 44100 5 8 2 4 +#generateq 44k_glass_space1 44100 5 8 2 8 +#generateq 44k_glass_space1 44100 5 8 2 16 +#generateq 44k_glass_space1 44100 5 8 2 32 +#generateq 44k_glass_space1 44100 5 8 2 2048 + diff --git a/glass_test3.sh b/glass_test3.sh new file mode 100644 index 0000000..029cf58 --- /dev/null +++ b/glass_test3.sh @@ -0,0 +1,112 @@ +function runq () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + qs=$4 + + exp_name="space_qs$dim" + + echo "__________________________________________" + echo "__________________________________________" + echo "__________________________________________" + echo "" + echo ">> running $exp_name" + echo "" + echo "__________________________________________" + echo "__________________________________________" + echo "__________________________________________" + + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels $qs \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 4 \ + --resume True +} + +function rundim () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + dim=$4 + + exp_name="space_dim$dim" + + echo "__________________________________________" + echo "__________________________________________" + echo "__________________________________________" + echo "" + echo ">> running $exp_name" + echo "" + echo "__________________________________________" + echo "__________________________________________" + echo "__________________________________________" + + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim $dim --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 8 \ + --resume True +} + +function runseq () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + seq=$4 + + exp_name="space_seq$seq" + + echo "__________________________________________" + echo "__________________________________________" + echo "__________________________________________" + echo "" + echo ">> running $exp_name" + echo "" + echo "__________________________________________" + echo "__________________________________________" + echo "__________________________________________" + + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len $seq --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 4 \ + --resume True +} + +#rundim 44k_glass_space1 44100 5 4096 +runseq 44k_glass_space1 44100 5 512 +rundim 44k_glass_space1 44100 5 512 +#rundim 44k_glass_space1 44100 5 256 +#rundim 44k_glass_space1 44100 5 128 +#rundim 44k_glass_space1 44100 5 64 +#rundim 44k_glass_space1 44100 5 32 +#rundim 44k_glass_space1 44100 5 16 +#rundim 44k_glass_space1 44100 5 8 +#rundim 44k_glass_space1 44100 5 4 +#rundim 44k_glass_space1 44100 5 2 +#rundim 44k_glass_space1 44100 5 1 + +runq 44k_glass_space1 44100 5 1024 +runq 44k_glass_space1 44100 5 128 + diff --git a/glass_tests.sh b/glass_tests.sh new file mode 100644 index 0000000..62d8932 --- /dev/null +++ b/glass_tests.sh @@ -0,0 +1,202 @@ +function run2 () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + fs1=$4 + fs2=$5 + + exp_name=space_2 + + echo "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@" + echo "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@" + echo "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@" + echo "" + echo ">> running $exp_name $4 $5" + echo "" + echo "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@" + echo "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@" + echo "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@" + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes $fs1 $fs2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 2 \ + --resume True +} +function generate2 () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + fs1=$4 + fs2=$5 + + exp_name=space_2 + + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo "" + echo ">> generating $exp_name $4 $5" + echo "" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + python generate.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes $fs1 $fs2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 1 \ + --n_samples 6 \ + --resume True +} +function run3 () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + fs1=$4 + fs2=$5 + fs3=$6 + + exp_name=space_3 + + echo ">> running $exp_name $4 $5 $6" + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes $fs1 $fs2 $fs3 \ + --n_rnn 3 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 1 \ + --resume True +} + +function run4 () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + fs1=$4 + fs2=$5 + fs3=$6 + fs4=$7 + + exp_name=space_4 + + echo ">> running $exp_name $4 $5 $6 $7" + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes $fs1 $fs2 $fs3 $fs4 \ + --n_rnn 4 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 4 \ + --resume True +} +function run5 () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + fs1=$4 + fs2=$5 + fs3=$6 + fs4=$7 + fs5=$8 + + exp_name=space_5 + + echo ">> running $exp_name $4 $5 $6 $7 $8" + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes $fs1 $fs2 $fs3 $fs4 $fs5 \ + --n_rnn 5 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 4 \ + --resume True +} +function run6 () { + dataset=$1 + sample_rate=$2 + duration=$3 + let sample_length=$2*$3 + fs1=$4 + fs2=$5 + fs3=$6 + fs4=$7 + fs5=$8 + fs6=$9 + + exp_name=space_6 + + echo ">> running $exp_name $4 $5 $6 $7 $8 $9" + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes $fs1 $fs2 $fs3 $fs4 $fs5 $fs6 \ + --n_rnn 5 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit 4 \ + --resume True +} + +run4 44k_glass_space1 44100 5 2 2 2 2 +run5 44k_glass_space1 44100 5 2 2 2 2 2 +run6 44k_glass_space1 44100 5 2 2 2 2 2 2 + +run4 44k_glass_space1 44100 5 8 2 2 2 +run5 44k_glass_space1 44100 5 8 2 2 2 2 +run6 44k_glass_space1 44100 5 8 2 2 2 2 2 +#generate2 44k_glass_space1 44100 5 8 2 +#run2 44k_glass_space1 44100 5 8 4 +#generate2 44k_glass_space1 44100 5 8 4 +#run2 44k_glass_space1 44100 5 8 1 +#generate2 44k_glass_space1 44100 5 8 1 +#run2 44k_glass_space1 44100 5 16 8 +#run2 44k_glass_space1 44100 5 16 4 +#run2 44k_glass_space1 44100 5 16 2 +#generate2 44k_glass_space1 44100 5 16 2 +#run2 44k_glass_space1 44100 5 16 1 +#generate2 44k_glass_space1 44100 5 16 1 + +#run2 44k_glass_space1 44100 5 4 2 +#run2 44k_glass_space1 44100 5 4 1 +#run2 44k_glass_space1 44100 5 2 1 + +#run3 44k_glass_space1 44100 5 8 2 1 +#run3 44k_glass_space1 44100 5 8 4 2 +#run3 44k_glass_space1 44100 5 16 8 2 +#run3 44k_glass_space1 44100 5 16 4 2 +#run3 44k_glass_space1 44100 5 16 2 1 + +#run2 44k_glass_space1 44100 5 6 2 +#run2 44k_glass_space1 44100 5 6 3 +#run3 44k_glass_space1 44100 5 5 3 2 +#run3 44k_glass_space1 44100 5 10 5 2 +#run3 44k_glass_space1 44100 5 8 5 2 + diff --git a/kick_test.sh b/kick_test.sh new file mode 100644 index 0000000..c3754a4 --- /dev/null +++ b/kick_test.sh @@ -0,0 +1,51 @@ +function generaterrr () { + exp_name=$1 + dataset=$2 + epoch_limit=$3 + sample_rate=$4 + duration=$5 + let sample_length=$4*$5 + qlev=$6 + + echo ">> generating $exp_name" + python generate.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels $qlev \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit $epoch_limit \ + --resume True +} +function runrrr () { + exp_name=$1 + dataset=$2 + epoch_limit=$3 + sample_rate=$4 + duration=$5 + let sample_length=$4*$5 + qlev=$6 + + echo ">> running $exp_name" + python train.py \ + --exp $exp_name --dataset $dataset \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels $qlev \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit $epoch_limit \ + --resume True +} + +runrrr kiq256 kiq 1 44100 1 256 +generaterrr kiq256 kiq 6 44100 1 256 + +runrrr kiq512 kiq 1 44100 1 512 +generaterrr kiq512 kiq 10 44100 1 512 + @@ -0,0 +1,20 @@ +#!/bin/bash + +if [ "$#" -ne 2 ]; then + echo "Usage: $0 <results/exp\:experiment> render_name" + exit +fi + +dir=$1 +name=$2 +now=`date +'%Y%m%d'` + +cd "$dir/samples" +mkdir "$name" +../../xfade.sh +cp mix.wav "../../../output/$name.wav" +mv *.wav "$name" +cd .. +vim log +cd ../.. + @@ -255,7 +255,7 @@ def main(exp, frame_sizes, dataset, **params): init_comet(params, trainer) - trainer.run(params['epoch_limit']) + trainer.run(int(params['epoch_limit'])) if __name__ == '__main__': diff --git a/train_basic_22k.sh b/train_basic_22k.sh new file mode 100644 index 0000000..5104034 --- /dev/null +++ b/train_basic_22k.sh @@ -0,0 +1,15 @@ +exp_name=$1 +dataset=$2 +epoch_limit=$3 +python train.py \ + --exp $exp_name --dataset $exp_name \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 64 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate 22050 \ + --sample_length 110250 \ + --keep_old_checkpoints False \ + --epoch_limit $epoch_limit \ + --resume True + diff --git a/train_basic_32k.sh b/train_basic_32k.sh new file mode 100644 index 0000000..c4d1319 --- /dev/null +++ b/train_basic_32k.sh @@ -0,0 +1,15 @@ +exp_name=$1 +dataset=$2 +epoch_limit=$3 +python train.py \ + --exp $exp_name --dataset $exp_name \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 64 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate 32000 \ + --sample_length 160000 \ + --keep_old_checkpoints False \ + --epoch_limit $epoch_limit \ + --resume True + diff --git a/train_basic_44k.sh b/train_basic_44k.sh new file mode 100644 index 0000000..1908f5d --- /dev/null +++ b/train_basic_44k.sh @@ -0,0 +1,15 @@ +exp_name=$1 +dataset=$2 +epoch_limit=$3 +python train.py \ + --exp $exp_name --dataset $exp_name \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 64 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate 44100 \ + --sample_length 220500 \ + --keep_old_checkpoints False \ + --epoch_limit $epoch_limit \ + --resume True + diff --git a/train_drums.sh b/train_drums.sh new file mode 100644 index 0000000..2157fc7 --- /dev/null +++ b/train_drums.sh @@ -0,0 +1,44 @@ +function generate () { + exp_name=$1 + epoch_limit=$2 + sample_rate=$3 + sample_length=$4 + + echo ">> generating $exp_name" + python generate.py \ + --exp $exp_name --dataset $exp_name \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit $epoch_limit \ + --resume True +} +function run () { + exp_name=$1 + epoch_limit=$2 + sample_rate=$3 + sample_length=$4 + + echo ">> running $exp_name" + python train.py \ + --exp $exp_name --dataset $exp_name \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --epoch_limit $epoch_limit \ + --resume True +} + +#generate 44k_jlin3 64 44100 4134 +generate 44k_jlin4 64 44100 4134 +generate 44k_clouds2 64 44100 4134 +generate 44k_glassmix 64 44100 4134 + diff --git a/train_test_generate.sh b/train_test_generate.sh new file mode 100644 index 0000000..af4f1c7 --- /dev/null +++ b/train_test_generate.sh @@ -0,0 +1,93 @@ +function generate () { + exp_name=$1 + n_samples=$2 + sample_rate=$3 + duration=$4 + let sample_length=$3*$4 + + echo "" + echo "###################################################" + echo "###################################################" + echo "###################################################" + echo "" + echo ">> generating $exp_name" + echo "" + echo "###################################################" + echo "###################################################" + echo "###################################################" + echo "" + python generate.py \ + --exp $exp_name --dataset $exp_name \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --keep_old_checkpoints False \ + --n_samples $n_samples \ + --epoch_limit 1 \ + --resume True +} +function run () { + exp_name=$1 + epoch_limit=$2 + n_samples=$3 + sample_rate=$4 + duration=$5 + let sample_length=$4*$5 + + echo "" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo "" + echo ">> running $exp_name" + echo "" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo ">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>" + echo "" + python train.py \ + --exp $exp_name --dataset $exp_name \ + --frame_sizes 8 2 \ + --n_rnn 2 --dim 1024 --q_levels 256 \ + --seq_len 1024 --batch_size 128 \ + --val_frac 0.1 --test_frac 0.1 \ + --sample_rate $sample_rate \ + --sample_length $sample_length \ + --n_samples $n_samples \ + --keep_old_checkpoints False \ + --epoch_limit $epoch_limit \ + --resume True +} +function standard () { + dataset=$1 + run $1 6 6 44100 5 +} +function quick () { + dataset=$1 + run $1 4 6 44100 5 +} +function fast () { + dataset=$1 + run $1 1 6 44100 10 +} + +standard 44k_blblbl2 +standard 44k_faty-scrub2 +standard 44k_faty-vocal2 + +quick 44k_siren +quick 44k_whatifvocode +quick 44k_jlin-faty + +fast 44k_lipnoise +quick 44k_jlin-faty +quick 44k_jlin-faty + +run 44k_dances 8 6 44100 5 +run 44k_jlin4 4 3 44100 10 + +run 44k_sundae 4 3 44100 10 + diff --git a/trainer/__init__.py b/trainer/__init__.py index 7e2ea18..1f39506 100644 --- a/trainer/__init__.py +++ b/trainer/__init__.py @@ -56,6 +56,15 @@ class Trainer(object): self.train() self.call_plugins('epoch', self.epochs) + def generate(self, epochs=1): + for q in self.plugin_queues.values(): + heapq.heapify(q) + + for self.epochs in range(self.epochs + 1, self.epochs + epochs + 1): + # self.train() + self.call_plugins('update', self.iterations, self.model) + self.call_plugins('epoch', self.epochs) + def train(self): for (self.iterations, data) in \ enumerate(self.dataset, self.iterations + 1): diff --git a/trainer/plugins.py b/trainer/plugins.py index f8c299b..0126870 100644 --- a/trainer/plugins.py +++ b/trainer/plugins.py @@ -141,7 +141,7 @@ class SaverPlugin(Plugin): class GeneratorPlugin(Plugin): - pattern = 'ep{}-s{}.wav' + pattern = 'd-{}-ep{}-s{}.wav' def __init__(self, samples_path, n_samples, sample_length, sample_rate): super().__init__([(1, 'epoch')]) @@ -159,7 +159,7 @@ class GeneratorPlugin(Plugin): for i in range(self.n_samples): write_wav( os.path.join( - self.samples_path, self.pattern.format(epoch_index, i + 1) + self.samples_path, self.pattern.format(int(time.time()), epoch_index, i + 1) ), samples[i, :], sr=self.sample_rate, norm=True ) @@ -168,7 +168,7 @@ class GeneratorPlugin(Plugin): class StatsPlugin(Plugin): data_file_name = 'stats.pkl' - plot_pattern = '{}.svg' + plot_pattern = 'd-{}-{}.svg' def __init__(self, results_path, iteration_fields, epoch_fields, plots): super().__init__([(1, 'iteration'), (1, 'epoch')]) @@ -252,7 +252,7 @@ class StatsPlugin(Plugin): pyplot.legend() pyplot.savefig( - os.path.join(self.results_path, self.plot_pattern.format(name)) + os.path.join(self.results_path, self.plot_pattern.format(int(time.time()), name)) ) @staticmethod |