Re-organize folder structure and hierarchy for DS2.

cloud/pcloud_submit.sh

@@ -1,7 +1,9 @@
-TRAIN_MANIFEST="cloud/cloud.manifest.train"
+#! /usr/bin/bash
-DEV_MANIFEST="cloud/cloud.manifest.dev"
+
+TRAIN_MANIFEST="cloud/cloud_manifests/cloud.manifest.train"
+DEV_MANIFEST="cloud/cloud_manifests/cloud.manifest.dev"
 CLOUD_MODEL_DIR="./checkpoints"
-BATCH_SIZE=256
+BATCH_SIZE=512
 NUM_GPU=8
 NUM_NODE=1
 IS_LOCAL="True"

cloud/pcloud_train.sh

@@ -1,3 +1,5 @@
+#! /usr/bin/bash
+
 TRAIN_MANIFEST=$1
 DEV_MANIFEST=$2
 MODEL_PATH=$3
@@ -14,11 +16,29 @@ python ./cloud/split_data.py \
 --out_manifest_path='/local.manifest.dev'
 
 python -u train.py \
---batch_size=$BATCH_SIZE \
+--batch_size=${BATCH_SIZE} \
---use_gpu=1 \
 --trainer_count=${NUM_GPU} \
---num_threads_data=${NUM_GPU} \
+--num_passes=200 \
+--num_proc_data=${NUM_GPU} \
+--num_conv_layers=2 \
+--num_rnn_layers=3 \
+--rnn_layer_size=2048 \
+--num_iter_print=100 \
+--learning_rate=5e-4 \
+--max_duration=27.0 \
+--min_duration=0.0 \
+--use_sortagrad=True \
+--use_gru=False \
+--use_gpu=True \
 --is_local=${IS_LOCAL} \
---train_manifest_path='/local.manifest.train' \
+--share_rnn_weights=True \
---dev_manifest_path='/local.manifest.dev' \
+--train_manifest='/local.manifest.train' \
---output_model_dir=${MODEL_PATH} 2>&1 | tee ./log/train.log
+--dev_manifest='/local.manifest.dev' \
+--mean_std_path='data/librispeech/mean_std.npz' \
+--vocab_path='data/librispeech/eng_vocab.txt' \
+--output_model_dir='./checkpoints' \
+--output_model_dir=${MODEL_PATH} \
+--augment_conf_path='conf/augmentation.config' \
+--specgram_type='linear' \
+--shuffle_method='batch_shuffle_clipped' \
+2>&1 | tee ./log/train.log

cloud/pcloud_upload_data.sh

@@ -1,5 +1,9 @@
-IN_MANIFESTS="../datasets/manifest.train ../datasets/manifest.dev ../datasets/manifest.test"
+#! /usr/bin/bash
-OUT_MANIFESTS="./cloud.manifest.train ./cloud.manifest.dev ./cloud.manifest.test"
+
+mkdir cloud_manifests
+
+IN_MANIFESTS="../data/librispeech/manifest.train ../data/librispeech/manifest.dev-clean ../data/librispeech/manifest.test-clean"
+OUT_MANIFESTS="cloud_manifests/cloud.manifest.train cloud_manifests/cloud.manifest.dev cloud_manifests/cloud.manifest.test"
 CLOUD_DATA_DIR="/pfs/dlnel/home/USERNAME/deepspeech2/data/librispeech"
 NUM_SHARDS=50
 
@@ -14,4 +18,5 @@ then
     echo "Upload Data Failed!"
     exit 1
 fi
+
 echo "All Done."

data_utils/augmentor/impulse_response.py

@@ -4,7 +4,7 @@ from __future__ import division
 from __future__ import print_function
 
 from data_utils.augmentor.base import AugmentorBase
-from data_utils import utils
+from data_utils.utility import read_manifest
 from data_utils.audio import AudioSegment
 
 
@@ -19,8 +19,7 @@ class ImpulseResponseAugmentor(AugmentorBase):
 
     def __init__(self, rng, impulse_manifest_path):
         self._rng = rng
-        self._impulse_manifest = utils.read_manifest(
+        self._impulse_manifest = read_manifest(impulse_manifest_path)
-            manifest_path=impulse_manifest_path)
 
     def transform_audio(self, audio_segment):
         """Add impulse response effect.

data_utils/augmentor/noise_perturb.py

@@ -4,7 +4,7 @@ from __future__ import division
 from __future__ import print_function
 
 from data_utils.augmentor.base import AugmentorBase
-from data_utils import utils
+from data_utils.utility import read_manifest
 from data_utils.audio import AudioSegment
 
 
@@ -25,8 +25,7 @@ class NoisePerturbAugmentor(AugmentorBase):
         self._min_snr_dB = min_snr_dB
         self._max_snr_dB = max_snr_dB
         self._rng = rng
-        self._noise_manifest = utils.read_manifest(
+        self._noise_manifest = read_manifest(manifest_path=noise_manifest_path)
-            manifest_path=noise_manifest_path)
 
     def transform_audio(self, audio_segment):
         """Add background noise audio.

data_utils/data.py

@@ -11,7 +11,7 @@ import multiprocessing
 import numpy as np
 import paddle.v2 as paddle
 from threading import local
-from data_utils import utils
+from data_utils.utility import read_manifest
 from data_utils.augmentor.augmentation import AugmentationPipeline
 from data_utils.featurizer.speech_featurizer import SpeechFeaturizer
 from data_utils.speech import SpeechSegment
@@ -159,7 +159,7 @@ class DataGenerator(object):
 
         def batch_reader():
             # read manifest
-            manifest = utils.read_manifest(
+            manifest = read_manifest(
                 manifest_path=manifest_path,
                 max_duration=self._max_duration,
                 min_duration=self._min_duration)

data_utils/featurizer/audio_featurizer.py

@@ -4,7 +4,7 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from data_utils import utils
+from data_utils.utility import read_manifest
 from data_utils.audio import AudioSegment
 from python_speech_features import mfcc
 from python_speech_features import delta

data_utils/normalizer.py

@@ -5,7 +5,7 @@ from __future__ import print_function
 
 import numpy as np
 import random
-import data_utils.utils as utils
+from data_utils.utility import read_manifest
 from data_utils.audio import AudioSegment
 
 
@@ -75,7 +75,7 @@ class FeatureNormalizer(object):
 
     def _compute_mean_std(self, manifest_path, featurize_func, num_samples):
         """Compute mean and std from randomly sampled instances."""
-        manifest = utils.read_manifest(manifest_path)
+        manifest = read_manifest(manifest_path)
         sampled_manifest = self._rng.sample(manifest, num_samples)
         features = []
         for instance in sampled_manifest:

datasets/run_all.sh

@@ -1,13 +0,0 @@
-cd librispeech
-python librispeech.py
-if [ $? -ne 0 ]; then
-    echo "Prepare LibriSpeech failed. Terminated."
-    exit 1
-fi
-cd -
-
-cat librispeech/manifest.train* | shuf > manifest.train
-cat librispeech/manifest.dev-clean > manifest.dev
-cat librispeech/manifest.test-clean > manifest.test
-
-echo "All done."

datasets/run_noise.sh

@@ -1,10 +0,0 @@
-cd noise 
-python chime3_background.py
-if [ $? -ne 0 ]; then
-    echo "Prepare CHiME3 background noise failed. Terminated."
-    exit 1
-fi
-cd -
-
-cat noise/manifest.* > manifest.noise
-echo "All done."

deploy/_init_paths.py

@@ -0,0 +1,19 @@
+"""Set up paths for DS2"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os.path
+import sys
+
+
+def add_path(path):
+    if path not in sys.path:
+        sys.path.insert(0, path)
+
+
+this_dir = os.path.dirname(__file__)
+
+# Add project path to PYTHONPATH
+proj_path = os.path.join(this_dir, '..')
+add_path(proj_path)

deploy/demo_server.py

@@ -9,10 +9,11 @@ import SocketServer
 import struct
 import wave
 import paddle.v2 as paddle
+import _init_paths
 from data_utils.data import DataGenerator
-from model import DeepSpeech2Model
+from models.model import DeepSpeech2Model
 from data_utils.utils import read_manifest
-from utils import add_arguments, print_arguments
+from utils.utility import add_arguments, print_arguments
 
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)
@@ -36,13 +37,13 @@ add_arg('speech_save_dir',  str,
         'demo_cache',
         "Directory to save demo audios.")
 add_arg('warmup_manifest',  str,
-        'datasets/manifest.test',
+        'data/librispeech/manifest.test-clean',
         "Filepath of manifest to warm up.")
 add_arg('mean_std_path',    str,
-        'mean_std.npz',
+        'data/librispeech/mean_std.npz',
         "Filepath of normalizer's mean & std.")
 add_arg('vocab_path',       str,
-        'datasets/vocab/eng_vocab.txt',
+        'data/librispeech/eng_vocab.txt',
         "Filepath of vocabulary.")
 add_arg('model_path',       str,
         './checkpoints/params.latest.tar.gz',

evaluate.py

@@ -7,9 +7,9 @@ import argparse
 import functools
 import paddle.v2 as paddle
 from data_utils.data import DataGenerator
-from model import DeepSpeech2Model
+from models.model import DeepSpeech2Model
-from error_rate import wer, cer
+from utils.error_rate import wer, cer
-from utils import add_arguments, print_arguments
+from utils.utility import add_arguments, print_arguments
 
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)
@@ -30,13 +30,13 @@ add_arg('use_gpu',          bool,   True,   "Use GPU or not.")
 add_arg('share_rnn_weights',bool,   True,   "Share input-hidden weights across "
                                             "bi-directional RNNs. Not for GRU.")
 add_arg('test_manifest',   str,
-        'datasets/manifest.test',
+        'data/librispeech/manifest.test-clean',
         "Filepath of manifest to evaluate.")
 add_arg('mean_std_path',    str,
-        'mean_std.npz',
+        'data/librispeech/mean_std.npz',
         "Filepath of normalizer's mean & std.")
 add_arg('vocab_path',       str,
-        'datasets/vocab/eng_vocab.txt',
+        'data/librispeech/eng_vocab.txt',
         "Filepath of vocabulary.")
 add_arg('model_path',       str,
         './checkpoints/params.latest.tar.gz',

examples/librispeech/generate.sh

@@ -0,0 +1,28 @@
+#! /usr/bin/bash
+
+pushd ../..
+
+CUDA_VISIBLE_DEVICES=0 \
+python -u infer.py \
+--num_samples=10 \
+--trainer_count=1 \
+--beam_size=500 \
+--num_proc_bsearch=12 \
+--num_proc_data=12 \
+--num_conv_layers=2 \
+--num_rnn_layers=3 \
+--rnn_layer_size=2048 \
+--alpha=0.36 \
+--beta=0.25 \
+--cutoff_prob=0.99 \
+--use_gru=False \
+--use_gpu=True \
+--share_rnn_weights=True \
+--infer_manifest='data/librispeech/manifest.dev-clean' \
+--mean_std_path='data/librispeech/mean_std.npz' \
+--vocab_path='data/librispeech/eng_vocab.txt' \
+--model_path='checkpoints/params.latest.tar.gz' \
+--lang_model_path='lm/data/common_crawl_00.prune01111.trie.klm' \
+--decoding_method='ctc_beam_search' \
+--error_rate_type='wer' \
+--specgram_type='linear'

examples/librispeech/prepare_data.sh

@@ -0,0 +1,32 @@
+#! /usr/bin/bash
+
+pushd ../..
+
+# download data, generate manifests
+python data/librispeech/librispeech.py \
+--manifest_prefix='data/librispeech/manifest' \
+--full_download='True' \
+--target_dir='~/.cache/paddle/dataset/speech/Libri'
+
+if [ $? -ne 0 ]; then
+    echo "Prepare LibriSpeech failed. Terminated."
+    exit 1
+fi
+
+#cat data/librispeech/manifest.train* | shuf > data/librispeech/manifest.train
+
+
+# compute mean and stddev for normalizer
+python tools/compute_mean_std.py \
+--manifest_path='data/librispeech/manifest.train' \
+--num_samples=2000 \
+--specgram_type='linear' \
+--output_path='data/librispeech/mean_std.npz'
+
+if [ $? -ne 0 ]; then
+    echo "Compute mean and stddev failed. Terminated."
+    exit 1
+fi
+
+
+echo "LibriSpeech Data preparation done."

examples/librispeech/run_test.sh

@@ -0,0 +1,28 @@
+#! /usr/bin/bash
+
+pushd ../..
+
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
+python -u evaluate.py \
+--batch_size=128 \
+--trainer_count=8 \
+--beam_size=500 \
+--num_proc_bsearch=12 \
+--num_proc_data=12 \
+--num_conv_layers=2 \
+--num_rnn_layers=3 \
+--rnn_layer_size=2048 \
+--alpha=0.36 \
+--beta=0.25 \
+--cutoff_prob=0.99 \
+--use_gru=False \
+--use_gpu=True \
+--share_rnn_weights=True \
+--test_manifest='data/librispeech/manifest.test-clean' \
+--mean_std_path='data/librispeech/mean_std.npz' \
+--vocab_path='data/librispeech/eng_vocab.txt' \
+--model_path='checkpoints/params.latest.tar.gz' \
+--lang_model_path='lm/data/common_crawl_00.prune01111.trie.klm' \
+--decoding_method='ctc_beam_search' \
+--error_rate_type='wer' \
+--specgram_type='linear'

examples/librispeech/run_train.sh

@@ -0,0 +1,30 @@
+#! /usr/bin/bash
+
+pushd ../..
+
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
+python -u train.py \
+--batch_size=256 \
+--trainer_count=8 \
+--num_passes=200 \
+--num_proc_data=12 \
+--num_conv_layers=2 \
+--num_rnn_layers=3 \
+--rnn_layer_size=2048 \
+--num_iter_print=100 \
+--learning_rate=5e-4 \
+--max_duration=27.0 \
+--min_duration=0.0 \
+--use_sortagrad=True \
+--use_gru=False \
+--use_gpu=True \
+--is_local=True \
+--share_rnn_weights=True \
+--train_manifest='data/librispeech/manifest.train' \
+--dev_manifest='data/librispeech/manifest.dev' \
+--mean_std_path='data/librispeech/mean_std.npz' \
+--vocab_path='data/librispeech/eng_vocab.txt' \
+--output_model_dir='./checkpoints' \
+--augment_conf_path='conf/augmentation.config' \
+--specgram_type='linear' \
+--shuffle_method='batch_shuffle_clipped'

examples/librispeech/run_tune.sh

@@ -0,0 +1,30 @@
+#! /usr/bin/bash
+
+pushd ../..
+
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
+python -u tools/tune.py \
+--num_samples=100 \
+--trainer_count=8 \
+--beam_size=500 \
+--num_proc_bsearch=12 \
+--num_conv_layers=2 \
+--num_rnn_layers=3 \
+--rnn_layer_size=2048 \
+--num_alphas=14 \
+--num_betas=20 \
+--alpha_from=0.1 \
+--alpha_to=0.36 \
+--beta_from=0.05 \
+--beta_to=1.0 \
+--cutoff_prob=0.99 \
+--use_gru=False \
+--use_gpu=True \
+--share_rnn_weights=True \
+--tune_manifest='data/librispeech/manifest.dev-clean' \
+--mean_std_path='data/librispeech/mean_std.npz' \
+--vocab_path='data/librispeech/eng_vocab.txt' \
+--model_path='checkpoints/params.latest.tar.gz' \
+--lang_model_path='lm/data/common_crawl_00.prune01111.trie.klm' \
+--error_rate_type='wer' \
+--specgram_type='linear'

infer.py

@@ -7,9 +7,9 @@ import argparse
 import functools
 import paddle.v2 as paddle
 from data_utils.data import DataGenerator
-from model import DeepSpeech2Model
+from models.model import DeepSpeech2Model
-from error_rate import wer, cer
+from utils.error_rate import wer, cer
-from utils import add_arguments, print_arguments
+from utils.utility import add_arguments, print_arguments
 
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)
@@ -29,13 +29,13 @@ add_arg('use_gpu',          bool,   True,   "Use GPU or not.")
 add_arg('share_rnn_weights',bool,   True,   "Share input-hidden weights across "
                                             "bi-directional RNNs. Not for GRU.")
 add_arg('infer_manifest',   str,
-        'datasets/manifest.dev',
+        'data/librispeech/manifest.dev-clean',
         "Filepath of manifest to infer.")
 add_arg('mean_std_path',    str,
-        'mean_std.npz',
+        'data/librispeech/mean_std.npz',
         "Filepath of normalizer's mean & std.")
 add_arg('vocab_path',       str,
-        'datasets/vocab/eng_vocab.txt',
+        'data/librispeech/eng_vocab.txt',
         "Filepath of vocabulary.")
 add_arg('lang_model_path',  str,
         'lm/data/common_crawl_00.prune01111.trie.klm',

models/model.py

@@ -7,10 +7,10 @@ import sys
 import os
 import time
 import gzip
-from decoder import *
-from lm.lm_scorer import LmScorer
 import paddle.v2 as paddle
-from layer import *
+from utils.decoder import ctc_greedy_decoder, ctc_beam_search_decoder
+from lm.lm_scorer import LmScorer
+from models.network import deep_speech_v2_network
 
 
 class DeepSpeech2Model(object):
@@ -241,7 +241,7 @@ class DeepSpeech2Model(object):
         text_data = paddle.layer.data(
             name="transcript_text",
             type=paddle.data_type.integer_value_sequence(vocab_size))
-        self._log_probs, self._loss = deep_speech2(
+        self._log_probs, self._loss = deep_speech_v2_network(
             audio_data=audio_data,
             text_data=text_data,
             dict_size=vocab_size,

models/network.py

@@ -1,4 +1,4 @@
-"""Contains DeepSpeech2 layers."""
+"""Contains DeepSpeech2 layers and networks."""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
@@ -205,16 +205,15 @@ def rnn_group(input, size, num_stacks, use_gru, share_rnn_weights):
     return output
 
 
-def deep_speech2(audio_data,
+def deep_speech_v2_network(audio_data,
-                 text_data,
+                           text_data,
-                 dict_size,
+                           dict_size,
-                 num_conv_layers=2,
+                           num_conv_layers=2,
-                 num_rnn_layers=3,
+                           num_rnn_layers=3,
-                 rnn_size=256,
+                           rnn_size=256,
-                 use_gru=False,
+                           use_gru=False,
-                 share_rnn_weights=True):
+                           share_rnn_weights=True):
-    """
+    """The DeepSpeech2 network structure.
-    The whole DeepSpeech2 model structure.
 
     :param audio_data: Audio spectrogram data layer.
     :type audio_data: LayerOutput

tools/build_vocab.py

@@ -13,8 +13,8 @@ import json
 from collections import Counter
 import os.path
 import _init_paths
-from data_utils import utils
+from data_utils.utility import read_manifest
-from utils import add_arguments, print_arguments
+from utils.utility import add_arguments, print_arguments
 
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)

tools/compute_mean_std.py

@@ -9,7 +9,7 @@ import _init_paths
 from data_utils.normalizer import FeatureNormalizer
 from data_utils.augmentor.augmentation import AugmentationPipeline
 from data_utils.featurizer.audio_featurizer import AudioFeaturizer
-from utils import add_arguments, print_arguments
+from utils.utility import add_arguments, print_arguments
 
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)

tools/tune.py

@@ -7,10 +7,11 @@ import numpy as np
 import argparse
 import functools
 import paddle.v2 as paddle
+import _init_paths
 from data_utils.data import DataGenerator
-from model import DeepSpeech2Model
+from models.model import DeepSpeech2Model
-from error_rate import wer
+from utils.error_rate import wer
-from utils import add_arguments, print_arguments
+from utils.utility import add_arguments, print_arguments
 
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)
@@ -27,20 +28,20 @@ add_arg('num_betas',        int,    20,     "# of beta candidates for tuning.")
 add_arg('alpha_from',       float,  0.1,    "Where alpha starts tuning from.")
 add_arg('alpha_to',         float,  0.36,   "Where alpha ends tuning with.")
 add_arg('beta_from',        float,  0.05,   "Where beta starts tuning from.")
-add_arg('beta_to',          float,  0.36,   "Where beta ends tuning with.")
+add_arg('beta_to',          float,  1.0,    "Where beta ends tuning with.")
 add_arg('cutoff_prob',      float,  0.99,   "Cutoff probability for pruning.")
 add_arg('use_gru',          bool,   False,  "Use GRUs instead of simple RNNs.")
 add_arg('use_gpu',          bool,   True,   "Use GPU or not.")
 add_arg('share_rnn_weights',bool,   True,   "Share input-hidden weights across "
                                             "bi-directional RNNs. Not for GRU.")
 add_arg('tune_manifest',    str,
-        'datasets/manifest.test',
+        'data/librispeech/manifest.dev',
         "Filepath of manifest to tune.")
 add_arg('mean_std_path',    str,
-        'mean_std.npz',
+        'data/librispeech/mean_std.npz',
         "Filepath of normalizer's mean & std.")
 add_arg('vocab_path',       str,
-        'datasets/vocab/eng_vocab.txt',
+        'data/librispeech/eng_vocab.txt',
         "Filepath of vocabulary.")
 add_arg('lang_model_path',  str,
         'lm/data/common_crawl_00.prune01111.trie.klm',

train.py

@@ -6,9 +6,9 @@ from __future__ import print_function
 import argparse
 import functools
 import paddle.v2 as paddle
-from model import DeepSpeech2Model
+from models.model import DeepSpeech2Model
 from data_utils.data import DataGenerator
-from utils import add_arguments, print_arguments
+from utils.utility import add_arguments, print_arguments
 
 parser = argparse.ArgumentParser(description=__doc__)
 add_arg = functools.partial(add_arguments, argparser=parser)
@@ -27,21 +27,21 @@ add_arg('max_duration',     float,  27.0,   "Longest audio duration allowed.")
 add_arg('min_duration',     float,  0.0,    "Shortest audio duration allowed.")
 add_arg('use_sortagrad',    bool,   True,   "Use SortaGrad or not.")
 add_arg('use_gpu',          bool,   True,   "Use GPU or not.")
-add_arg('is_local',         bool,   True,   "Use pserver or not.")
 add_arg('use_gru',          bool,   False,  "Use GRUs instead of simple RNNs.")
+add_arg('is_local',         bool,   True,   "Use pserver or not.")
 add_arg('share_rnn_weights',bool,   True,   "Share input-hidden weights across "
                                             "bi-directional RNNs. Not for GRU.")
 add_arg('train_manifest',   str,
-        'datasets/manifest.train',
+        'data/librispeech/manifest.train',
         "Filepath of train manifest.")
 add_arg('dev_manifest',     str,
-        'datasets/manifest.dev',
+        'data/librispeech/manifest.dev-clean',
         "Filepath of validation manifest.")
 add_arg('mean_std_path',    str,
-        'mean_std.npz',
+        'data/librispeech/mean_std.npz',
         "Filepath of normalizer's mean & std.")
 add_arg('vocab_path',       str,
-        'datasets/vocab/eng_vocab.txt',
+        'data/librispeech/eng_vocab.txt',
         "Filepath of vocabulary.")
 add_arg('init_model_path',  str,
         None,
@@ -101,7 +101,7 @@ def train():
         rnn_layer_size=args.rnn_layer_size,
         use_gru=args.use_gru,
         pretrained_model_path=args.init_model_path,
-        share_rnn_weights=args.share_weights)
+        share_rnn_weights=args.share_rnn_weights)
     ds2_model.train(
         train_batch_reader=train_batch_reader,
         dev_batch_reader=dev_batch_reader,