fix data

deepspeech/exps/deepspeech2/bin/deploy/runtime.py

@@ -168,7 +168,7 @@ if __name__ == "__main__":
                         default=False,
                         help="Whether use gpu.")
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()
@@ -180,7 +180,7 @@ if __name__ == "__main__":
     print(config)
 
     args.warmup_manifest = config.data.test_manifest
-    print_arguments(args)
+    print_arguments(args, globals())
 
     if args.dump_config:
         with open(args.dump_config, 'w') as f:

deepspeech/exps/deepspeech2/bin/deploy/server.py

@@ -98,7 +98,7 @@ if __name__ == "__main__":
             "Directory to save demo audios.")
     add_arg('warmup_manifest', str, None, "Filepath of manifest to warm up.")
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()
@@ -110,7 +110,7 @@ if __name__ == "__main__":
     print(config)
 
     args.warmup_manifest = config.data.test_manifest
-    print_arguments(args)
+    print_arguments(args, globals())
 
     if args.dump_config:
         with open(args.dump_config, 'w') as f:

deepspeech/exps/deepspeech2/bin/infer.py

@@ -33,7 +33,7 @@ def main(config, args):
 if __name__ == "__main__":
     parser = default_argument_parser()
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()

deepspeech/exps/deepspeech2/bin/test.py

@@ -32,7 +32,7 @@ def main(config, args):
 if __name__ == "__main__":
     parser = default_argument_parser()
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()

deepspeech/exps/deepspeech2/bin/train.py

@@ -37,7 +37,7 @@ def main(config, args):
 if __name__ == "__main__":
     parser = default_argument_parser()
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()

deepspeech/exps/deepspeech2/bin/tune.py

@@ -168,7 +168,7 @@ if __name__ == "__main__":
     add_arg('cutoff_top_n', int, 40, "Cutoff number for pruning.")
 
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()

deepspeech/exps/u2/bin/export.py

@@ -33,7 +33,7 @@ def main(config, args):
 if __name__ == "__main__":
     parser = default_argument_parser()
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()

deepspeech/exps/u2/bin/test.py

@@ -34,7 +34,7 @@ def main(config, args):
 if __name__ == "__main__":
     parser = default_argument_parser()
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()

deepspeech/exps/u2/bin/train.py

@@ -38,7 +38,7 @@ def main(config, args):
 if __name__ == "__main__":
     parser = default_argument_parser()
     args = parser.parse_args()
-    print_arguments(args)
+    print_arguments(args, globals())
 
     # https://yaml.org/type/float.html
     config = get_cfg_defaults()

deepspeech/frontend/featurizer/text_featurizer.py

@@ -34,11 +34,12 @@ class TextFeaturizer(object):
         """
         assert unit_type in ('char', 'spm', 'word')
         self.unit_type = unit_type
-        self._vocab_dict, self._id2token, self._vocab_list = self._load_vocabulary_from_file(
-            vocab_filepath)
         self.unk = UNK
-        self.unk_id = self._vocab_list.index(self.unk)
-        self.eos_id = self._vocab_list.index(EOS)
+        if vocab_filepath:
+            self._vocab_dict, self._id2token, self._vocab_list = self._load_vocabulary_from_file(
+                vocab_filepath)
+            self.unk_id = self._vocab_list.index(self.unk)
+            self.eos_id = self._vocab_list.index(EOS)
 
         if unit_type == 'spm':
             spm_model = spm_model_prefix + '.model'

deepspeech/utils/utility.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 """Contains common utility functions."""
 
+import os
 import math
 import distutils.util
 from typing import List
@@ -20,7 +21,7 @@ from typing import List
 __all__ = ['print_arguments', 'add_arguments', "log_add"]
 
 
-def print_arguments(args):
+def print_arguments(args, info=None):
     """Print argparse's arguments.
 
     Usage:
@@ -35,10 +36,14 @@ def print_arguments(args):
     :param args: Input argparse.Namespace for printing.
     :type args: argparse.Namespace
     """
-    print("-----------  Configuration Arguments -----------")
+    filename = ""
+    if info:
+        filename = info["__file__"]
+    filename = os.path.basename(filename)
+    print(f"----------- {filename} Configuration Arguments -----------")
     for arg, value in sorted(vars(args).items()):
         print("%s: %s" % (arg, value))
-    print("------------------------------------------------")
+    print("-----------------------------------------------------------")
 
 
 def add_arguments(argname, type, default, help, argparser, **kwargs):

examples/aishell/README.md

@@ -1 +1,2 @@
 * s0 for deepspeech2
+* s1 for u2

examples/aishell/s0/local/data.sh

@@ -1,12 +1,11 @@
 #! /usr/bin/env bash
 
 mkdir -p data
-
 TARGET_DIR=${MAIN_ROOT}/examples/dataset
 mkdir -p ${TARGET_DIR}
 
 # download data, generate manifests
-PYTHONPATH=.:$PYTHONPATH python3 ${TARGET_DIR}/aishell/aishell.py \
+python3 ${TARGET_DIR}/aishell/aishell.py \
 --manifest_prefix="data/manifest" \
 --target_dir="${TARGET_DIR}/aishell"
 
@@ -16,11 +15,17 @@ if [ $? -ne 0 ]; then
 fi
 
 
+for dataset in train dev test; do
+    mv data/manifest.${dataset} data/manifest.${dataset}.raw
+done
+
+
 # build vocabulary
 python3 ${MAIN_ROOT}/utils/build_vocab.py \
+--unit_type="char" \
 --count_threshold=0 \
 --vocab_path="data/vocab.txt" \
---manifest_paths "data/manifest.train" "data/manifest.dev"
+--manifest_paths "data/manifest.train.raw" "data/manifest.dev.raw"
 
 if [ $? -ne 0 ]; then
     echo "Build vocabulary failed. Terminated."
@@ -30,9 +35,11 @@ fi
 
 # compute mean and stddev for normalizer
 python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
---manifest_path="data/manifest.train" \
+--manifest_path="data/manifest.train.raw" \
 --num_samples=2000 \
---specgram_type="linear" \
+--specgram_type="fbank" \
+--feat_dim=80 \
+--delta_delta=false \
 --output_path="data/mean_std.npz"
 
 if [ $? -ne 0 ]; then
@@ -41,5 +48,21 @@ if [ $? -ne 0 ]; then
 fi
 
 
+# format manifest with tokenids, vocab size
+for dataset in train dev test; do
+    python3 ${MAIN_ROOT}/utils/format_data.py \
+    --feat_type "raw" \
+    --cmvn_path "data/mean_std.npz" \
+    --unit_type "char" \
+    --vocab_path="data/vocab.txt" \
+    --manifest_path="data/manifest.${dataset}.raw" \
+    --output_path="data/manifest.${dataset}"
+done
+
+if [ $? -ne 0 ]; then
+    echo "Formt mnaifest failed. Terminated."
+    exit 1
+fi
+
 echo "Aishell data preparation done."
 exit 0

examples/aishell/s0/local/download_model.sh

@@ -1,22 +0,0 @@
-#! /usr/bin/env bash
-
-. ${MAIN_ROOT}/utils/utility.sh
-
-DIR=data/pretrain
-mkdir -p ${DIR} 
-
-URL='https://deepspeech.bj.bcebos.com/mandarin_models/aishell_model_fluid.tar.gz'
-MD5=2bf0cc8b6d5da2a2a787b5cc36a496b5
-TARGET=${DIR}/aishell_model_fluid.tar.gz
-
-
-echo "Download Aishell model ..."
-download $URL $MD5 $TARGET
-if [ $? -ne 0 ]; then
-    echo "Fail to download Aishell model!"
-    exit 1
-fi
-tar -zxvf $TARGET -C ${DIR}
-
-
-exit 0

examples/aishell/s1/conf/augmentation.config

@@ -0,0 +1,8 @@
+[
+    {
+        "type": "shift",
+        "params": {"min_shift_ms": -5,
+                   "max_shift_ms": 5},
+        "prob": 1.0
+    }
+]

examples/aishell/s1/local/data.sh

@@ -0,0 +1 @@
+../../s0/local/data.sh

examples/aishell/s1/local/download_lm_ch.sh

@@ -0,0 +1 @@
+../../s0/local/download_lm_ch.sh

examples/aishell/s1/local/export.sh

@@ -0,0 +1,20 @@
+#! /usr/bin/env bash
+
+if [ $# != 2 ];then
+    echo "usage: export ckpt_path jit_model_path"
+    exit -1
+fi
+
+python3 -u ${BIN_DIR}/export.py \
+--config conf/deepspeech2.yaml \
+--checkpoint_path ${1} \
+--export_path ${2} 
+
+
+if [ $? -ne 0 ]; then
+    echo "Failed in evaluation!"
+    exit 1
+fi
+
+
+exit 0

examples/aishell/s1/local/test.sh

@@ -0,0 +1,21 @@
+#! /usr/bin/env bash
+
+# download language model
+bash local/download_lm_ch.sh
+if [ $? -ne 0 ]; then
+    exit 1
+fi
+
+python3 -u ${BIN_DIR}/test.py \
+--device 'gpu' \
+--nproc 1 \
+--config conf/deepspeech2.yaml \
+--output ckpt
+
+if [ $? -ne 0 ]; then
+    echo "Failed in evaluation!"
+    exit 1
+fi
+
+
+exit 0

examples/aishell/s1/local/train.sh

@@ -0,0 +1,23 @@
+#! /usr/bin/env bash
+
+# train model
+# if you wish to resume from an exists model, uncomment --init_from_pretrained_model
+export FLAGS_sync_nccl_allreduce=0
+
+ngpu=$(echo ${CUDA_VISIBLE_DEVICES} | python -c 'import sys; a = sys.stdin.read(); print(len(a.split(",")));')
+echo "using $ngpu gpus..."
+
+python3 -u ${BIN_DIR}/train.py \
+--device 'gpu' \
+--nproc ${ngpu} \
+--config conf/deepspeech2.yaml \
+--output ckpt-${1}
+
+
+if [ $? -ne 0 ]; then
+    echo "Failed in training!"
+    exit 1
+fi
+
+
+exit 0

examples/aishell/s1/path.sh

@@ -0,0 +1,14 @@
+export MAIN_ROOT=${PWD}/../../../
+
+export PATH=${MAIN_ROOT}:${PWD}/tools:${PATH}
+export LC_ALL=C
+
+# Use UTF-8 in Python to avoid UnicodeDecodeError when LC_ALL=C
+export PYTHONIOENCODING=UTF-8 
+export PYTHONPATH=${MAIN_ROOT}:${PYTHONPATH}
+
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib/
+
+
+MODEL=u2
+export BIN_DIR=${MAIN_ROOT}/deepspeech/exps/${MODEL}/bin

examples/aishell/s1/run.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+
+source path.sh
+# only demos
+
+# prepare data
+bash ./local/data.sh
+
+# train model
+CUDA_VISIBLE_DEVICES=0,1,2,3 bash ./local/train.sh
+
+# test model
+CUDA_VISIBLE_DEVICES=0 bash ./local/test.sh
+
+# infer model
+CUDA_VISIBLE_DEVICES=0 bash ./local/infer.sh ckpt/checkpoints/step-3284
+
+# export model
+bash ./local/export.sh ckpt/checkpoints/step-3284 jit.model

examples/tiny/s0/local/data.sh

@@ -5,7 +5,7 @@ TARGET_DIR=${MAIN_ROOT}/examples/dataset
 mkdir -p ${TARGET_DIR}
 
 # download data, generate manifests
-PYTHONPATH=.:$PYTHONPATH python3 ${TARGET_DIR}/librispeech/librispeech.py \
+python3 ${TARGET_DIR}/librispeech/librispeech.py \
 --manifest_prefix="data/manifest" \
 --target_dir="${TARGET_DIR}/librispeech" \
 --full_download="False"
@@ -24,7 +24,7 @@ bpeprefix="data/bpe_${bpemode}_${nbpe}"
 # build vocabulary
 python3 ${MAIN_ROOT}/utils/build_vocab.py \
 --unit_type "spm" \
---vocab_size=${nbpe} \
+--spm_vocab_size=${nbpe} \
 --spm_mode ${bpemode} \
 --spm_model_prefix ${bpeprefix} \
 --vocab_path="data/vocab.txt" \

utils/build_vocab.py

@@ -45,9 +45,9 @@ add_arg('manifest_paths', str,
         nargs='+',
         required=True)
 # bpe
-add_arg('vocab_size', int, 0, "Vocab size for spm.")
+add_arg('spm_vocab_size', int, 0, "Vocab size for spm.")
 add_arg('spm_mode', str, 'unigram', "spm model type, e.g. unigram, spm, char, word. only need when `unit_type` is spm")
-add_arg('spm_model_prefix', str, "spm_model_%(spm_mode)_%(count_threshold)", "spm model prefix, only need when `unit_type` is spm")
+add_arg('spm_model_prefix', str, "", "spm_model_%(spm_mode)_%(count_threshold), spm model prefix, only need when `unit_type` is spm")
 # yapf: disable
 args = parser.parse_args()
 
@@ -64,7 +64,7 @@ def dump_text_manifest(fileobj, manifest_path):
         fileobj.write(line_json['text'] + "\n")
 
 def main():
-    print_arguments(args)
+    print_arguments(args, globals())
 
     fout = open(args.vocab_path, 'w', encoding='utf-8')
     fout.write(BLANK + "\n")  # 0 will be used for "blank" in CTC
@@ -91,7 +91,7 @@ def main():
         os.unlink(fp.name)
 
     # encode
-    text_feature = TextFeaturizer(args.unit_type, args.vocab_path, args.spm_model_prefix)
+    text_feature = TextFeaturizer(args.unit_type, "", args.spm_model_prefix)
     counter = Counter()
 
     for manifest_path in args.manifest_paths:

utils/compute_mean_std.py

@@ -46,7 +46,7 @@ args = parser.parse_args()
 
 
 def main():
-    print_arguments(args)
+    print_arguments(args, globals())
 
     augmentation_pipeline = AugmentationPipeline('{}')
     audio_featurizer = AudioFeaturizer(

utils/format_data.py

@@ -48,12 +48,12 @@ args = parser.parse_args()
 
 
 def main():
-    print_arguments(args)
+    print_arguments(args, globals())
     fout = open(args.output_path, 'w', encoding='utf-8')
 
     # get feat dim
     mean, std = load_cmvn(args.cmvn_path, filetype='npz')
-    feat_dim = mean.shape[0]
+    feat_dim = mean.shape[1] #(1, D)
     print(f"Feature dim: {feat_dim}")
 
     text_feature = TextFeaturizer(args.unit_type, args.vocab_path, args.spm_model_prefix)