Hugging Face's logo

Spaces:
akhaliq
/
VQMIVC
Runtime error

App Files Community
2
VQMIVC
File size: 8,888 Bytes 
2b7bf83
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
 
#!/bin/bash

# Copyright 2020 Tomoki Hayashi
#  MIT License (https://opensource.org/licenses/MIT)

. ./cmd.sh || exit 1;
. ./path.sh || exit 1;

# basic settings
stage=0        # stage to start
stop_stage=100 # stage to stop
verbose=1      # verbosity level (lower is less info)
n_gpus=1       # number of gpus in training
n_jobs=4       # number of parallel jobs in feature extraction

# NOTE(kan-bayashi): renamed to conf to avoid conflict in parse_options.sh
conf=conf/parallel_wavegan.v1.yaml

# directory path setting
db_root=/path/to/database # direcotry including spk name directory (MODIFY BY YOURSELF)
                          # e.g.
                          # /path/to/database
                          # ├── spk_1
                          # │   ├── utt1.wav
                          # ├── spk_2
                          # │   ├── utt1.wav
                          # │   ...
                          # └── spk_N
                          #     ├── utt1.wav
                          #     ...
dumpdir=dump # directory to dump features

# subset setting
spks="all"    # speaker name to be used (e.g. "spk1 spk2")
              # it must be matched the name under the ${db_root}
              # if set to "all", all of the speakers in ${db_root} will be used
shuffle=false # whether to shuffle the data to create subset
num_dev=10    # the number of development data for each speaker
num_eval=10   # the number of evaluation data for each speaker
              # (if set to 0, the same dev set is used as eval set)

# training related setting
tag=""     # tag for directory to save model
resume=""  # checkpoint path to resume training
           # (e.g. <path>/<to>/checkpoint-10000steps.pkl)
pretrain="" # checkpoint path to load pretrained parameters
            # (e.g. ../../jsut/<path>/<to>/checkpoint-400000steps.pkl)

# decoding related setting
checkpoint="" # checkpoint path to be used for decoding
              # if not provided, the latest one will be used
              # (e.g. <path>/<to>/checkpoint-400000steps.pkl)

# shellcheck disable=SC1091
. utils/parse_options.sh || exit 1;

train_set="train_nodev_$(echo "${spks}" | tr " " "_")" # name of training data directory
dev_set="dev_$(echo "${spks}" | tr " " "_")"           # name of development data directory
eval_set="eval_$(echo "${spks}" | tr " " "_")"         # name of evaluation data directory

set -euo pipefail

if [ "${stage}" -le 0 ] && [ "${stop_stage}" -ge 0 ]; then
    echo "Stage 0: Data preparation"
    train_data_dirs=""
    dev_data_dirs=""
    eval_data_dirs=""
    if [ "${spks}" = "all" ]; then
        spks=$(find "${db_root}" -maxdepth 1 ! -path "${db_root}" \
            -follow -type d -print0 -name "[^.]*" | xargs -0 -I{} basename {})
    fi
    for spk in ${spks}; do
        local/data_prep.sh \
            --fs "$(yq ".sampling_rate" "${conf}")" \
            --shuffle "${shuffle}" \
            --num_dev "${num_dev}" \
            --num_eval "${num_eval}" \
            --train_set "train_nodev_${spk}" \
            --dev_set "dev_${spk}" \
            --eval_set "eval_${spk}" \
            "${db_root}" "${spk}" data
        train_data_dirs+=" data/train_nodev_${spk}"
        dev_data_dirs+=" data/dev_${spk}"
        eval_data_dirs+=" data/eval_${spk}"
    done
    # shellcheck disable=SC2086
    utils/combine_data.sh "data/${train_set}" ${train_data_dirs}
    # shellcheck disable=SC2086
    utils/combine_data.sh "data/${dev_set}" ${dev_data_dirs}
    # shellcheck disable=SC2086
    utils/combine_data.sh "data/${eval_set}" ${eval_data_dirs}
fi

stats_ext=$(grep -q "hdf5" <(yq ".format" "${conf}") && echo "h5" || echo "npy")
if [ "${stage}" -le 1 ] && [ "${stop_stage}" -ge 1 ]; then
    echo "Stage 1: Feature extraction"
    # extract raw features
    pids=()
    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
    (
        [ ! -e "${dumpdir}/${name}/raw" ] && mkdir -p "${dumpdir}/${name}/raw"
        echo "Feature extraction start. See the progress via ${dumpdir}/${name}/raw/preprocessing.*.log."
        utils/make_subset_data.sh "data/${name}" "${n_jobs}" "${dumpdir}/${name}/raw"
        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/raw/preprocessing.JOB.log" \
            parallel-wavegan-preprocess \
                --config "${conf}" \
                --scp "${dumpdir}/${name}/raw/wav.JOB.scp" \
                --dumpdir "${dumpdir}/${name}/raw/dump.JOB" \
                --verbose "${verbose}"
        echo "Successfully finished feature extraction of ${name} set."
    ) &
    pids+=($!)
    done
    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
    echo "Successfully finished feature extraction."

    # calculate statistics for normalization
    if [ -z "${pretrain}" ]; then
        # calculate statistics for normalization
        echo "Statistics computation start. See the progress via ${dumpdir}/${train_set}/compute_statistics.log."
        ${train_cmd} "${dumpdir}/${train_set}/compute_statistics.log" \
            parallel-wavegan-compute-statistics \
                --config "${conf}" \
                --rootdir "${dumpdir}/${train_set}/raw" \
                --dumpdir "${dumpdir}/${train_set}" \
                --verbose "${verbose}"
        echo "Successfully finished calculation of statistics."
    else
        echo "Use statistics of pretrained model. Skip statistics computation."
        cp "$(dirname "${pretrain}")/stats.${stats_ext}" "${dumpdir}/${train_set}"
    fi

    # normalize and dump them
    pids=()
    for name in "${train_set}" "${dev_set}" "${eval_set}"; do
    (
        [ ! -e "${dumpdir}/${name}/norm" ] && mkdir -p "${dumpdir}/${name}/norm"
        echo "Nomalization start. See the progress via ${dumpdir}/${name}/norm/normalize.*.log."
        ${train_cmd} JOB=1:${n_jobs} "${dumpdir}/${name}/norm/normalize.JOB.log" \
            parallel-wavegan-normalize \
                --config "${conf}" \
                --stats "${dumpdir}/${train_set}/stats.${stats_ext}" \
                --rootdir "${dumpdir}/${name}/raw/dump.JOB" \
                --dumpdir "${dumpdir}/${name}/norm/dump.JOB" \
                --verbose "${verbose}"
        echo "Successfully finished normalization of ${name} set."
    ) &
    pids+=($!)
    done
    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
    echo "Successfully finished normalization."
fi

if [ -z "${tag}" ]; then
    expdir="exp/${train_set}_$(basename "${conf}" .yaml)"
    if [ -n "${pretrain}" ]; then
        pretrain_tag=$(basename "$(dirname "${pretrain}")")
        expdir+="_${pretrain_tag}"
    fi
else
    expdir="exp/${train_set}_${tag}"
fi
if [ "${stage}" -le 2 ] && [ "${stop_stage}" -ge 2 ]; then
    echo "Stage 2: Network training"
    [ ! -e "${expdir}" ] && mkdir -p "${expdir}"
    cp "${dumpdir}/${train_set}/stats.${stats_ext}" "${expdir}"
    if [ "${n_gpus}" -gt 1 ]; then
        train="python -m parallel_wavegan.distributed.launch --nproc_per_node ${n_gpus} -c parallel-wavegan-train"
    else
        train="parallel-wavegan-train"
    fi
    echo "Training start. See the progress via ${expdir}/train.log."
    ${cuda_cmd} --gpu "${n_gpus}" "${expdir}/train.log" \
        ${train} \
            --config "${conf}" \
            --train-dumpdir "${dumpdir}/${train_set}/norm" \
            --dev-dumpdir "${dumpdir}/${dev_set}/norm" \
            --outdir "${expdir}" \
            --resume "${resume}" \
            --pretrain "${pretrain}" \
            --verbose "${verbose}"
    echo "Successfully finished training."
fi

if [ "${stage}" -le 3 ] && [ "${stop_stage}" -ge 3 ]; then
    echo "Stage 3: Network decoding"
    # shellcheck disable=SC2012
    [ -z "${checkpoint}" ] && checkpoint="$(ls -dt "${expdir}"/*.pkl | head -1 || true)"
    outdir="${expdir}/wav/$(basename "${checkpoint}" .pkl)"
    pids=()
    for name in "${dev_set}" "${eval_set}"; do
    (
        [ ! -e "${outdir}/${name}" ] && mkdir -p "${outdir}/${name}"
        [ "${n_gpus}" -gt 1 ] && n_gpus=1
        echo "Decoding start. See the progress via ${outdir}/${name}/decode.log."
        ${cuda_cmd} --gpu "${n_gpus}" "${outdir}/${name}/decode.log" \
            parallel-wavegan-decode \
                --dumpdir "${dumpdir}/${name}/norm" \
                --checkpoint "${checkpoint}" \
                --outdir "${outdir}/${name}" \
                --verbose "${verbose}"
        echo "Successfully finished decoding of ${name} set."
    ) &
    pids+=($!)
    done
    i=0; for pid in "${pids[@]}"; do wait "${pid}" || ((++i)); done
    [ "${i}" -gt 0 ] && echo "$0: ${i} background jobs are failed." && exit 1;
    echo "Successfully finished decoding."
fi
echo "Finished."