Spaces:
Running
on
Zero
Running
on
Zero
File size: 4,133 Bytes
115cf00 90433f5 115cf00 7555afb 115cf00 |
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 |
#!/bin/bash
# Copyright 2024 Alibaba Inc. All Rights Reserved.
. ./path.sh || exit 1;
stage=-1
stop_stage=3
data_url=www.openslr.org/resources/68
data_dir=/mnt/hengwu.zty/data/tts/openslr/magicdata-read
pretrained_model_dir=../../../pretrained_models/CosyVoice-300M
if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
echo "Data Download"
for part in dev_set test_set train_set; do
local/download_and_untar.sh ${data_dir} ${data_url} ${part}
done
fi
if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
echo "Data preparation, prepare wav.scp/text/utt2spk/spk2utt"
for x in dev test train; do
mkdir -p data/$x
python local/prepare_data.py --src_dir $data_dir/$x --des_dir data/$x
done
fi
if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
echo "Extract campplus speaker embedding, you will get spk2embedding.pt and utt2embedding.pt in data/$x dir"
for x in dev test train; do
tools/extract_embedding.py --dir data/$x \
--onnx_path $pretrained_model_dir/campplus.onnx
done
fi
if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
echo "Extract discrete speech token, you will get utt2speech_token.pt in data/$x dir"
for x in dev test train; do
tools/extract_speech_token.py --dir data/$x \
--onnx_path $pretrained_model_dir/speech_tokenizer_v1.onnx
done
fi
if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
echo "Prepare required parquet format data, you should have prepared wav.scp/text/utt2spk/spk2utt/utt2embedding.pt/spk2embedding.pt/utt2speech_token.pt"
for x in dev test train; do
mkdir -p data/$x/parquet
tools/make_parquet_list.py --num_utts_per_parquet 1000 \
--num_processes 10 \
--src_dir data/$x \
--des_dir data/$x/parquet
done
fi
# inference
if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
echo "Run inference. Please make sure utt in tts_text is in prompt_data"
for mode in sft zero_shot; do
python cosyvoice/bin/inference.py --mode $mode \
--gpu 0 \
--config conf/cosyvoice.yaml \
--prompt_data data/test/parquet/data.list \
--prompt_utt2data data/test/parquet/utt2data.list \
--tts_text `pwd`/tts_text.json \
--llm_model $pretrained_model_dir/llm.pt \
--flow_model $pretrained_model_dir/flow.pt \
--hifigan_model $pretrained_model_dir/hift.pt \
--result_dir `pwd`/exp/cosyvoice/test/$mode
done
fi
# train llm
export CUDA_VISIBLE_DEVICES="0,1,2,3"
num_gpus=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
job_id=1986
dist_backend="nccl"
num_workers=2
prefetch=100
train_engine=torch_ddp
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
echo "Run train. We only support llm traning for now. If your want to train from scratch, please use conf/cosyvoice.fromscratch.yaml"
if [ $train_engine == 'deepspeed' ]; then
echo "Notice deepspeed has its own optimizer config. Modify conf/ds_stage2.json if necessary"
fi
cp data/train/parquet/data.list data/train.data.list
cp data/dev/parquet/data.list data/dev.data.list
for model in llm flow; do
torchrun --nnodes=1 --nproc_per_node=$num_gpus \
--rdzv_id=$job_id --rdzv_backend="c10d" --rdzv_endpoint="localhost:0" \
cosyvoice/bin/train.py \
--train_engine $train_engine \
--config conf/cosyvoice.yaml \
--train_data data/train.data.list \
--cv_data data/dev.data.list \
--model $model \
--checkpoint $pretrained_model_dir/$model.pt \
--model_dir `pwd`/exp/cosyvoice/$model/$train_engine \
--tensorboard_dir `pwd`/tensorboard/cosyvoice/$model/$train_engine \
--ddp.dist_backend $dist_backend \
--num_workers ${num_workers} \
--prefetch ${prefetch} \
--pin_memory \
--deepspeed_config ./conf/ds_stage2.json \
--deepspeed.save_states model+optimizer
done
fi
if [ ${stage} -le 6 ] && [ ${stop_stage} -ge 6 ]; then
echo "Export your model for inference speedup. Remember copy your llm or flow model to model_dir"
python cosyvoice/bin/export_jit.py --model_dir $pretrained_model_dir
python cosyvoice/bin/export_onnx.py --model_dir $pretrained_model_dir
fi |