Hugging Face's logo

Spaces:
shenyunhang
/
VITA-Audio
Running on Zero

App Files Community
VITA-Audio
File size: 13,522 Bytes 
52e4f53
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
 
#!/bin/bash

set -e
set -x

SEQ_LENGTH="$1"
if [ -z "$SEQ_LENGTH" ]
then
	SEQ_LENGTH=32768
fi

timestamp="$2"
if [ -z "$timestamp" ]
then
	timestamp=`date +'%Y%m%d_%H%M%S'`
fi

######################################################################
export ROOT_PATH=/data/
export CODE_PATH=${ROOT_PATH}/VITA-Audio/

export LOCAL_ROOT_PATH=/data_local/
export LOCAL_CODE_PATH=${LOCAL_ROOT_PATH}/VITA-Audio/
mkdir -p ${LOCAL_ROOT_PATH}
mkdir -p ${LOCAL_CODE_PATH}

apt install -y rsync
mkdir -p ${LOCAL_CODE_PATH}
rsync -a --exclude ".git" --exclude ".gitee" ${CODE_PATH}/ ${LOCAL_CODE_PATH}/

cd ${LOCAL_CODE_PATH}
rm -fr datasets
ln -s ${ROOT_PATH}/data datasets

######################################################################
SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
source ${CODE_PATH}/scripts/set_env_ds_gpu.sh
pip3 install transformers==4.48.3
#pip3 install --no-index --find-links=/data/software/ transformers==4.48.3

######################################################################
OUTPUT_DIR=${ROOT_PATH}/output/LM/"$0"/${timestamp}/

mkdir -p ${OUTPUT_DIR}
rsync -avh $0 ${OUTPUT_DIR}

export HF_HOME="${ROOT_PATH}/data/HF_HOME/"
mkdir -p ${HF_HOME}
export HF_ENDPOINT=https://hf-mirror.com

export MODELSCOPE_CACHE="${ROOT_PATH}/data/MODELSCOPE_CACHE/"
mkdir -p ${MODELSCOPE_CACHE}

export LC_ALL="en_US.utf8"

######################################################################
LOG=${OUTPUT_DIR}/log_node${INDEX}.txt
exec &> >(tee -a "$LOG")
echo Logging output to "$LOG"


######################################################################
if true
#if false
then
	MODEL_NAME_OR_PATH="/data/output/LM/scripts/deepspeed/sts_qwen25/finetune_glm4voice_mtp10_stage2.sh/VITA-Audio-Boost/"
	MODEL_NAME_OR_PATH="/data/output/LM/scripts/deepspeed/sts_qwen25/finetune_glm4voice_mtp10_stage2.sh/VITA-Audio-Balance/"

	AUDIO_TOKENIZER_PATH=${ROOT_PATH}/models/THUDM/glm-4-voice-tokenizer
	FLOW_PATH=${ROOT_PATH}/models/THUDM/glm-4-voice-decoder
	AUDIO_TOKENIZER_TYPE="glm4voice"

	export PYTHONPATH=${PYTHONPATH}:${LOCAL_CODE_PATH}/third_party/GLM-4-Voice/:${LOCAL_CODE_PATH}/third_party/GLM-4-Voice/cosyvoice/:${LOCAL_CODE_PATH}/third_party/GLM-4-Voice/third_party/Matcha-TTS/

fi

######################################################################
DISTRIBUTED_ARGS="
--nproc_per_node $NPROC_PER_NODE \
	--nnodes $NNODES \
	--node_rank $NODE_RANK \
	--master_addr $MASTER_ADDR \
	--master_port $MASTER_PORT
	"

######################################################################
if true
#if false
then
	apt-get update && apt install -y ffmpeg

	JSON_PATH=${ROOT_PATH}/data/jsonl/fixie-ai/llama-questions/test.jsonl
	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_sqa.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/llama-questions/

	python evaluation/compute-acc-of-contain.py ${OUTPUT_DIR}/llama-questions/test_hyp_ref_text.json
	echo "copypaste ACC: ${JSON_PATH}"
	python evaluation/compute-acc-of-contain.py ${OUTPUT_DIR}/llama-questions/test_hyp_ref_speech.json
	echo "copypaste ACC: ${JSON_PATH}"


	JSON_PATH=${ROOT_PATH}/data/jsonl/fixie-ai/trivia_qa-audio/validation.jsonl
	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_sqa.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/trivia_qa-audio/

	python evaluation/compute-acc-of-contain.py ${OUTPUT_DIR}/trivia_qa-audio/validation_hyp_ref_text.json
	echo "copypaste ACC: ${JSON_PATH}"
	python evaluation/compute-acc-of-contain.py ${OUTPUT_DIR}/trivia_qa-audio/validation_hyp_ref_speech.json
	echo "copypaste ACC: ${JSON_PATH}"


	JSON_PATH=${ROOT_PATH}/data/jsonl/fixie-ai/spoken-web-questions/test.jsonl
	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_sqa.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/spoken-web-questions/

	python evaluation/compute-acc-of-contain.py ${OUTPUT_DIR}/spoken-web-questions/test_hyp_ref_text.json
	echo "copypaste ACC: ${JSON_PATH}"
	python evaluation/compute-acc-of-contain.py ${OUTPUT_DIR}/spoken-web-questions/test_hyp_ref_speech.json
	echo "copypaste ACC: ${JSON_PATH}"

fi


######################################################################
if true
#if false
then
	JSON_PATH=${ROOT_PATH}/data/jsonl/fixie-ai/librispeech_asr/validation.clean.jsonl

	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_asr.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/librispeech_asr/

	#python evaluation/compute-cer.py --char=1 --v=1 ${OUTPUT_DIR}/librispeech_asr/validation.clean_ref.txt ${OUTPUT_DIR}/librispeech_asr/validation.clean_hyp.txt
	#echo "copypaste CER: ${JSON_PATH}"
	python evaluation/compute-wer.py --char=1 --v=1 ${OUTPUT_DIR}/librispeech_asr/validation.clean_ref.txt ${OUTPUT_DIR}/librispeech_asr/validation.clean_hyp.txt
	echo "copypaste WER: ${JSON_PATH}"

	JSON_PATH=${ROOT_PATH}/data/jsonl/fixie-ai/librispeech_asr/validation.other.jsonl

	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_asr.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/librispeech_asr/

	#python evaluation/compute-cer.py --char=1 --v=1 ${OUTPUT_DIR}/librispeech_asr/validation.other_ref.txt ${OUTPUT_DIR}/librispeech_asr/validation.other_hyp.txt
	#echo "copypaste CER: ${JSON_PATH}"
	python evaluation/compute-wer.py --char=1 --v=1 ${OUTPUT_DIR}/librispeech_asr/validation.other_ref.txt ${OUTPUT_DIR}/librispeech_asr/validation.other_hyp.txt
	echo "copypaste WER: ${JSON_PATH}"


	JSON_PATH=${ROOT_PATH}/data/jsonl/fixie-ai/librispeech_asr/test.clean.jsonl

	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_asr.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/librispeech_asr/

	#python evaluation/compute-cer.py --char=1 --v=1 ${OUTPUT_DIR}/librispeech_asr/test.clean_ref.txt ${OUTPUT_DIR}/librispeech_asr/test.clean_hyp.txt
	#echo "copypaste CER: ${JSON_PATH}"
	python evaluation/compute-wer.py --char=1 --v=1 ${OUTPUT_DIR}/librispeech_asr/test.clean_ref.txt ${OUTPUT_DIR}/librispeech_asr/test.clean_hyp.txt
	echo "copypaste WER: ${JSON_PATH}"

	JSON_PATH=${ROOT_PATH}/data/jsonl/fixie-ai/librispeech_asr/test.other.jsonl

	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_asr.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/librispeech_asr/

	#python evaluation/compute-cer.py --char=1 --v=1 ${OUTPUT_DIR}/librispeech_asr/test.other_ref.txt ${OUTPUT_DIR}/librispeech_asr/test.other_hyp.txt
	#echo "copypaste CER: ${JSON_PATH}"
	python evaluation/compute-wer.py --char=1 --v=1 ${OUTPUT_DIR}/librispeech_asr/test.other_ref.txt ${OUTPUT_DIR}/librispeech_asr/test.other_hyp.txt
	echo "copypaste WER: ${JSON_PATH}"

fi


######################################################################
if true
#if false
then
	JSON_PATH=${ROOT_PATH}/data/jsonl/wenet-e2e/wenetspeech/TEST_MEETING.jsonl
	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_asr.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/wenetspeech/

	python evaluation/compute-cer.py --char=1 --v=1 ${OUTPUT_DIR}/wenetspeech/TEST_MEETING_ref.txt ${OUTPUT_DIR}/wenetspeech/TEST_MEETING_hyp.txt
	echo "copypaste CER: ${JSON_PATH}"
	python evaluation/compute-wer.py --char=1 --v=1 ${OUTPUT_DIR}/wenetspeech/TEST_MEETING_ref.txt ${OUTPUT_DIR}/wenetspeech/TEST_MEETING_hyp.txt
	echo "copypaste WER: ${JSON_PATH}"

	JSON_PATH=${ROOT_PATH}/data/jsonl/wenet-e2e/wenetspeech/TEST_NET.jsonl
	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_asr.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/wenetspeech/

	python evaluation/compute-cer.py --char=1 --v=1 ${OUTPUT_DIR}/wenetspeech/TEST_NET_ref.txt ${OUTPUT_DIR}/wenetspeech/TEST_NET_hyp.txt
	echo "copypaste CER: ${JSON_PATH}"
	python evaluation/compute-wer.py --char=1 --v=1 ${OUTPUT_DIR}/wenetspeech/TEST_NET_ref.txt ${OUTPUT_DIR}/wenetspeech/TEST_NET_hyp.txt
	echo "copypaste WER: ${JSON_PATH}"
fi


######################################################################
if true
#if false
then
	JSON_PATH=${ROOT_PATH}/data/jsonl/shenyunhang/AISHELL-1/test.jsonl

	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_asr.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/AISHELL-1/

	#python evaluation/compute-cer.py --char=1 --v=1 ${OUTPUT_DIR}/AISHELL-1/_test.clean_ref.txt ${OUTPUT_DIR}/AISHELL-1/test.clean_hyp.txt
	#echo "copypaste CER: ${JSON_PATH}"
	python evaluation/compute-wer.py --char=1 --v=1 ${OUTPUT_DIR}/AISHELL-1/test_ref.txt ${OUTPUT_DIR}/AISHELL-1/test_hyp.txt
	echo "copypaste WER: ${JSON_PATH}"


fi


######################################################################
if true
#if false
then
	JSON_PATH=${ROOT_PATH}/data/jsonl/mythicinfinity/libritts/test.clean.jsonl
	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_libritts.py \
		--json_path ${JSON_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/libritts/ \

	#python evaluation/compute-cer.py --char=1 --v=1 ${OUTPUT_DIR}/libritts/test.clean_ref.txt ${OUTPUT_DIR}/libritts/test.clean_hyp.txt
	#echo "copypaste CER: ${JSON_PATH}"
	python evaluation/compute-wer.py --char=1 --v=1 ${OUTPUT_DIR}/libritts/test.clean_ref.txt ${OUTPUT_DIR}/libritts/test.clean_hyp.txt
	echo "copypaste WER: ${JSON_PATH}"
fi


######################################################################
if true
#if false
then

	DATA_PATH=${ROOT_PATH}/data/BytedanceSpeech/seed-tts-eval/
	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_seedtts.py \
		--data_path ${DATA_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/seed-tts/ \
		--speaker_prompt \

	export ARNOLD_WORKER_GPU=${NPROC_PER_NODE}
	cd ${LOCAL_CODE_PATH}/third_party/seed-tts-eval

	bash cal_wer.sh ${DATA_PATH}/seedtts_testset/zh/meta.lst ${OUTPUT_DIR}/seed-tts/zh/ zh
	echo "copypaste WER: ${DATA_PATH} zh"
	bash cal_wer.sh ${DATA_PATH}/seedtts_testset/zh/hardcase.lst ${OUTPUT_DIR}/seed-tts/hardcase/ zh
	echo "copypaste WER: ${DATA_PATH} hardcase"
	bash cal_wer.sh ${DATA_PATH}/seedtts_testset/en/meta.lst ${OUTPUT_DIR}/seed-tts/en/ en
	echo "copypaste WER: ${DATA_PATH} en"

	bash cal_sim.sh ${DATA_PATH}/seedtts_testset/zh/meta.lst ${OUTPUT_DIR}/seed-tts/zh/ ${DATA_PATH}/wavlm_large_finetune.pth
	echo "copypaste SIM: ${DATA_PATH} zh"
	bash cal_sim.sh ${DATA_PATH}/seedtts_testset/zh/hardcase.lst ${OUTPUT_DIR}/seed-tts/hardcase/ ${DATA_PATH}/wavlm_large_finetune.pth
	echo "copypaste SIM: ${DATA_PATH} hardcase"
	bash cal_sim.sh ${DATA_PATH}/seedtts_testset/en/meta.lst ${OUTPUT_DIR}/seed-tts/en/ ${DATA_PATH}/wavlm_large_finetune.pth
	echo "copypaste SIM: ${DATA_PATH} en"

	cd ${LOCAL_CODE_PATH}

fi


######################################################################
if false
then
	DATA_PATH=${ROOT_PATH}/data/BytedanceSpeech/seed-tts-eval/
	torchrun ${DISTRIBUTED_ARGS} evaluation/evaluate_seedtts.py \
		--data_path ${DATA_PATH} \
		--model_name_or_path ${MODEL_NAME_OR_PATH} \
		--audio_tokenizer_path ${AUDIO_TOKENIZER_PATH} \
		--audio_tokenizer_type ${AUDIO_TOKENIZER_TYPE} \
		--flow_path ${FLOW_PATH} \
		--output_dir ${OUTPUT_DIR}/seed-tts/ \

	export ARNOLD_WORKER_GPU=${NPROC_PER_NODE}
	cd ${LOCAL_CODE_PATH}/third_party/seed-tts-eval

	bash cal_wer.sh ${DATA_PATH}/seedtts_testset/zh/meta.lst ${OUTPUT_DIR}/seed-tts/zh/ zh
	echo "copypaste WER: ${DATA_PATH} zh"
	bash cal_wer.sh ${DATA_PATH}/seedtts_testset/zh/hardcase.lst ${OUTPUT_DIR}/seed-tts/hardcase/ zh
	echo "copypaste WER: ${DATA_PATH} hardcase"
	bash cal_wer.sh ${DATA_PATH}/seedtts_testset/en/meta.lst ${OUTPUT_DIR}/seed-tts/en/ en
	echo "copypaste WER: ${DATA_PATH} en"

	cd ${LOCAL_CODE_PATH}

fi


set +x