提交 472cf70e 编写于 作者: H Hui Zhang

refactor egs; add utils; add tools; rm notebook;add speechnn; more docs;

上级 5ef4a34e
unset GREP_OPTIONS
# https://zhuanlan.zhihu.com/p/33050965
alias nvs='nvidia-smi'
alias his='history'
alias jobs='jobs -l'
alias ports='netstat -tulanp'
alias wget='wget -c'
## Colorize the grep command output for ease of use (good for log files)##
alias grep='grep --color=auto'
alias egrep='egrep --color=auto'
alias fgrep='fgrep --color=auto'
...@@ -42,6 +42,10 @@ ignore = ...@@ -42,6 +42,10 @@ ignore =
# these ignores are from flake8-comprehensions; please fix! # these ignores are from flake8-comprehensions; please fix!
C400,C401,C402,C403,C404,C405,C407,C411,C413,C414,C415 C400,C401,C402,C403,C404,C405,C407,C411,C413,C414,C415
per-file-ignores =
*/__init__.py: F401
# Specify the list of error codes you wish Flake8 to report. # Specify the list of error codes you wish Flake8 to report.
select = select =
E, E,
......
...@@ -10,8 +10,15 @@ ...@@ -10,8 +10,15 @@
.ipynb_checkpoints .ipynb_checkpoints
*.npz *.npz
*.done *.done
*.whl
tools/venv tools/venv
tools/kenlm tools/kenlm
tools/sox-14.4.2 tools/sox-14.4.2
tools/soxbindings tools/soxbindings
tools/montreal-forced-aligner/
tools/Montreal-Forced-Aligner/
tools/sctk
tools/sctk-20159b5/
*output/
...@@ -87,3 +87,9 @@ pull_request_rules: ...@@ -87,3 +87,9 @@ pull_request_rules:
actions: actions:
label: label:
add: ["Docker"] add: ["Docker"]
- name: "auto add label=Deployment"
conditions:
- files~=^speechnn/
actions:
label:
add: ["Deployment"]
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "academic-surname",
"metadata": {},
"outputs": [],
"source": [
"import paddle\n",
"from paddle import nn"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "fundamental-treasure",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/workspace/DeepSpeech-2.x/tools/venv-dev/lib/python3.7/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n",
" and should_run_async(code)\n"
]
}
],
"source": [
"L = nn.Linear(256, 2048)\n",
"L2 = nn.Linear(2048, 256)"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "consolidated-elephant",
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"import torch\n"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "moderate-noise",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"float64\n",
"Tensor(shape=[2, 51, 256], dtype=float32, place=CUDAPlace(0), stop_gradient=True,\n",
" [[[-1.54171216, -2.61531472, -1.79881978, ..., -0.31395876, 0.56513089, -0.44516513],\n",
" [-0.79492962, 1.91157901, 0.66567147, ..., 0.54825783, -1.01471853, -0.84924090],\n",
" [-1.22556651, -0.36225814, 0.65063190, ..., 0.65726501, 0.05563191, 0.09009409],\n",
" ...,\n",
" [ 0.38615900, -0.77905393, 0.99732304, ..., -1.38463700, -3.32365036, -1.31089687],\n",
" [ 0.05579993, 0.06885809, -1.66662002, ..., -0.23346378, -3.29372883, 1.30561364],\n",
" [ 1.90676069, 1.95093191, -0.28849599, ..., -0.06860496, 0.95347673, 1.00475824]],\n",
"\n",
" [[-0.91453546, 0.55298805, -1.06146812, ..., -0.86378336, 1.00454640, 1.26062179],\n",
" [ 0.10223761, 0.81301165, 2.36865163, ..., 0.16821407, 0.29240361, 1.05408621],\n",
" [-1.33196676, 1.94433689, 0.01934209, ..., 0.48036841, 0.51585966, 1.22893548],\n",
" ...,\n",
" [-0.19558455, -0.47075930, 0.90796155, ..., -1.28598249, -0.24321797, 0.17734711],\n",
" [ 0.89819717, -1.39516675, 0.17138045, ..., 2.39761519, 1.76364994, -0.52177650],\n",
" [ 0.94122332, -0.18581429, 1.36099780, ..., 0.67647684, -0.04699665, 1.51205540]]])\n",
"tensor([[[-1.5417, -2.6153, -1.7988, ..., -0.3140, 0.5651, -0.4452],\n",
" [-0.7949, 1.9116, 0.6657, ..., 0.5483, -1.0147, -0.8492],\n",
" [-1.2256, -0.3623, 0.6506, ..., 0.6573, 0.0556, 0.0901],\n",
" ...,\n",
" [ 0.3862, -0.7791, 0.9973, ..., -1.3846, -3.3237, -1.3109],\n",
" [ 0.0558, 0.0689, -1.6666, ..., -0.2335, -3.2937, 1.3056],\n",
" [ 1.9068, 1.9509, -0.2885, ..., -0.0686, 0.9535, 1.0048]],\n",
"\n",
" [[-0.9145, 0.5530, -1.0615, ..., -0.8638, 1.0045, 1.2606],\n",
" [ 0.1022, 0.8130, 2.3687, ..., 0.1682, 0.2924, 1.0541],\n",
" [-1.3320, 1.9443, 0.0193, ..., 0.4804, 0.5159, 1.2289],\n",
" ...,\n",
" [-0.1956, -0.4708, 0.9080, ..., -1.2860, -0.2432, 0.1773],\n",
" [ 0.8982, -1.3952, 0.1714, ..., 2.3976, 1.7636, -0.5218],\n",
" [ 0.9412, -0.1858, 1.3610, ..., 0.6765, -0.0470, 1.5121]]])\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"/workspace/DeepSpeech-2.x/tools/venv-dev/lib/python3.7/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n",
" and should_run_async(code)\n"
]
}
],
"source": [
"x = np.random.randn(2, 51, 256)\n",
"print(x.dtype)\n",
"px = paddle.to_tensor(x, dtype='float32')\n",
"tx = torch.tensor(x, dtype=torch.float32)\n",
"print(px)\n",
"print(tx)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cooked-progressive",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 5,
"id": "mechanical-prisoner",
"metadata": {},
"outputs": [],
"source": [
"data = np.load('enc_0_ff_out.npz', allow_pickle=True)\n",
"t_norm_ff = data['norm_ff']\n",
"t_ff_out = data['ff_out']\n",
"t_ff_l_x = data['ff_l_x']\n",
"t_ff_l_a_x = data['ff_l_a_x']\n",
"t_ff_l_a_l_x = data['ff_l_a_l_x']\n",
"t_ps = data['ps']"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "indie-marriage",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 6,
"id": "assured-zambia",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"True\n",
"True\n",
"True\n",
"True\n"
]
}
],
"source": [
"L.set_state_dict({'weight': t_ps[0].T, 'bias': t_ps[1]})\n",
"L2.set_state_dict({'weight': t_ps[2].T, 'bias': t_ps[3]})\n",
"\n",
"ps = []\n",
"for n, p in L.named_parameters():\n",
" ps.append(p)\n",
"\n",
"for n, p in L2.state_dict().items():\n",
" ps.append(p)\n",
" \n",
"for p, tp in zip(ps, t_ps):\n",
" print(np.allclose(p.numpy(), tp.T))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "committed-jacob",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "extreme-traffic",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "optimum-milwaukee",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 7,
"id": "viral-indian",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"True\n",
"True\n",
"True\n",
"True\n"
]
}
],
"source": [
"# data = np.load('enc_0_ff_out.npz', allow_pickle=True)\n",
"# t_norm_ff = data['norm_ff']\n",
"# t_ff_out = data['ff_out']\n",
"# t_ff_l_x = data['ff_l_x']\n",
"# t_ff_l_a_x = data['ff_l_a_x']\n",
"# t_ff_l_a_l_x = data['ff_l_a_l_x']\n",
"# t_ps = data['ps']\n",
"TL = torch.nn.Linear(256, 2048)\n",
"TL2 = torch.nn.Linear(2048, 256)\n",
"TL.load_state_dict({'weight': torch.tensor(t_ps[0]), 'bias': torch.tensor(t_ps[1])})\n",
"TL2.load_state_dict({'weight': torch.tensor(t_ps[2]), 'bias': torch.tensor(t_ps[3])})\n",
"\n",
"# for n, p in TL.named_parameters():\n",
"# print(n, p)\n",
"# for n, p in TL2.named_parameters():\n",
"# print(n, p)\n",
"\n",
"ps = []\n",
"for n, p in TL.state_dict().items():\n",
" ps.append(p.data.numpy())\n",
" \n",
"for n, p in TL2.state_dict().items():\n",
" ps.append(p.data.numpy())\n",
" \n",
"for p, tp in zip(ps, t_ps):\n",
" print(np.allclose(p, tp))"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "skilled-vietnamese",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[[[ 0.67277956 0.08313607 -0.62761104 ... -0.17480263 0.42718208\n",
" -0.5787626 ]\n",
" [ 0.91516656 0.5393416 1.7159258 ... 0.06144593 0.06486575\n",
" -0.03350811]\n",
" [ 0.438351 0.6227843 0.24096036 ... 1.0912522 -0.90929437\n",
" -1.012989 ]\n",
" ...\n",
" [ 0.68631977 0.14240924 0.10763275 ... -0.11513516 0.48065388\n",
" 0.04070369]\n",
" [-0.9525228 0.23197874 0.31264272 ... 0.5312439 0.18773697\n",
" -0.8450228 ]\n",
" [ 0.42024016 -0.04561988 0.54541194 ... -0.41933843 -0.00436018\n",
" -0.06663495]]\n",
"\n",
" [[-0.11638781 -0.33566502 -0.20887226 ... 0.17423287 -0.9195841\n",
" -0.8161046 ]\n",
" [-0.3469874 0.88269687 -0.11887559 ... -0.15566081 0.16357468\n",
" -0.20766167]\n",
" [-0.3847657 0.3984318 -0.06963477 ... -0.00360622 1.2360432\n",
" -0.26811332]\n",
" ...\n",
" [ 0.08230796 -0.46158582 0.54582864 ... 0.15747628 -0.44790155\n",
" 0.06020184]\n",
" [-0.8095085 0.43163058 -0.42837143 ... 0.8627463 0.90656304\n",
" 0.15847842]\n",
" [-1.485811 -0.18216592 -0.8882585 ... 0.32596245 0.7822631\n",
" -0.6460344 ]]]\n",
"[[[ 0.67278004 0.08313602 -0.6276114 ... -0.17480245 0.42718196\n",
" -0.5787625 ]\n",
" [ 0.91516703 0.5393413 1.7159253 ... 0.06144581 0.06486579\n",
" -0.03350812]\n",
" [ 0.43835106 0.62278455 0.24096027 ... 1.0912521 -0.9092943\n",
" -1.0129892 ]\n",
" ...\n",
" [ 0.6863195 0.14240888 0.10763284 ... -0.11513527 0.48065376\n",
" 0.04070365]\n",
" [-0.9525231 0.23197863 0.31264275 ... 0.53124386 0.18773702\n",
" -0.84502304]\n",
" [ 0.42024007 -0.04561983 0.545412 ... -0.41933888 -0.00436005\n",
" -0.066635 ]]\n",
"\n",
" [[-0.11638767 -0.33566508 -0.20887226 ... 0.17423296 -0.9195838\n",
" -0.8161046 ]\n",
" [-0.34698725 0.88269705 -0.11887549 ... -0.15566081 0.16357464\n",
" -0.20766166]\n",
" [-0.3847657 0.3984319 -0.06963488 ... -0.00360619 1.2360426\n",
" -0.26811326]\n",
" ...\n",
" [ 0.08230786 -0.4615857 0.5458287 ... 0.15747619 -0.44790167\n",
" 0.06020182]\n",
" [-0.8095083 0.4316307 -0.42837155 ... 0.862746 0.9065631\n",
" 0.15847899]\n",
" [-1.485811 -0.18216613 -0.8882584 ... 0.32596254 0.7822631\n",
" -0.6460344 ]]]\n",
"True\n",
"False\n"
]
}
],
"source": [
"y = L(px)\n",
"print(y.numpy())\n",
"\n",
"ty = TL(tx)\n",
"print(ty.data.numpy())\n",
"print(np.allclose(px.numpy(), tx.detach().numpy()))\n",
"print(np.allclose(y.numpy(), ty.detach().numpy()))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "incorrect-allah",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "prostate-cameroon",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 9,
"id": "governmental-surge",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[[ 0.04476918 0.554463 -0.3027508 ... -0.49600336 0.3751858\n",
" 0.8254095 ]\n",
" [ 0.95594174 -0.29528382 -1.2899452 ... 0.43718258 0.05584608\n",
" -0.06974669]]\n",
"[[ 0.04476918 0.5544631 -0.3027507 ... -0.49600336 0.37518573\n",
" 0.8254096 ]\n",
" [ 0.95594174 -0.29528376 -1.2899454 ... 0.4371827 0.05584623\n",
" -0.0697467 ]]\n",
"True\n",
"False\n",
"True\n"
]
}
],
"source": [
"x = np.random.randn(2, 256)\n",
"px = paddle.to_tensor(x, dtype='float32')\n",
"tx = torch.tensor(x, dtype=torch.float32)\n",
"y = L(px)\n",
"print(y.numpy())\n",
"ty = TL(tx)\n",
"print(ty.data.numpy())\n",
"print(np.allclose(px.numpy(), tx.detach().numpy()))\n",
"print(np.allclose(y.numpy(), ty.detach().numpy()))\n",
"print(np.allclose(y.numpy(), ty.detach().numpy(), atol=1e-5))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "confidential-jacket",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 10,
"id": "improved-civilization",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"5e7e7c9fde8350084abf1898cf52651cfc84b17a\n"
]
}
],
"source": [
"print(paddle.version.commit)"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "d1e2d3b4",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['__builtins__',\n",
" '__cached__',\n",
" '__doc__',\n",
" '__file__',\n",
" '__loader__',\n",
" '__name__',\n",
" '__package__',\n",
" '__spec__',\n",
" 'commit',\n",
" 'full_version',\n",
" 'istaged',\n",
" 'major',\n",
" 'minor',\n",
" 'mkl',\n",
" 'patch',\n",
" 'rc',\n",
" 'show',\n",
" 'with_mkl']"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"dir(paddle.version)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "c880c719",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"2.1.0\n"
]
}
],
"source": [
"print(paddle.version.full_version)"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "f26977bf",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"commit: 5e7e7c9fde8350084abf1898cf52651cfc84b17a\n",
"None\n"
]
}
],
"source": [
"print(paddle.version.show())"
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "04ad47f6",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"1.6.0\n"
]
}
],
"source": [
"print(torch.__version__)"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "e1e03830",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['__builtins__',\n",
" '__cached__',\n",
" '__doc__',\n",
" '__file__',\n",
" '__loader__',\n",
" '__name__',\n",
" '__package__',\n",
" '__spec__',\n",
" '__version__',\n",
" 'cuda',\n",
" 'debug',\n",
" 'git_version',\n",
" 'hip']"
]
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"dir(torch.version)"
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "4ad0389b",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'b31f58de6fa8bbda5353b3c77d9be4914399724d'"
]
},
"execution_count": 19,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"torch.version.git_version"
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "7870ea10",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'10.2'"
]
},
"execution_count": 21,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"torch.version.cuda"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "db8ee5a7",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "6321ec2a",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.0"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
此差异已折叠。
此差异已折叠。
此差异已折叠。
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "breeding-haven",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"/home/ssd5/zhanghui/DeepSpeech2.x\n"
]
},
{
"data": {
"text/plain": [
"'/home/ssd5/zhanghui/DeepSpeech2.x'"
]
},
"execution_count": 1,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"%cd ..\n",
"%pwd"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "appropriate-theta",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"LICENSE deepspeech examples\t\t requirements.txt tools\r\n",
"README.md docs\t libsndfile-1.0.28\t setup.sh\t utils\r\n",
"README_cn.md env.sh\t libsndfile-1.0.28.tar.gz tests\r\n"
]
}
],
"source": [
"!ls"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "entire-bloom",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/home/ssd5/zhanghui/DeepSpeech2.x/tools/venv/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.\n",
"Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
" def convert_to_list(value, n, name, dtype=np.int):\n",
"WARNING:root:override cat of paddle.Tensor if exists or register, remove this when fixed!\n",
"WARNING:root:register user masked_fill to paddle.Tensor, remove this when fixed!\n",
"WARNING:root:register user masked_fill_ to paddle.Tensor, remove this when fixed!\n",
"WARNING:root:register user repeat to paddle.Tensor, remove this when fixed!\n",
"WARNING:root:register user glu to paddle.nn.functional, remove this when fixed!\n",
"WARNING:root:register user GLU to paddle.nn, remove this when fixed!\n",
"WARNING:root:register user ConstantPad2d to paddle.nn, remove this when fixed!\n",
"WARNING:root:override ctc_loss of paddle.nn.functional if exists, remove this when fixed!\n"
]
}
],
"source": [
"from deepspeech.modules import loss"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "governmental-aircraft",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/home/ssd5/zhanghui/DeepSpeech2.x/tools/venv/lib/python3.7/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n",
" and should_run_async(code)\n"
]
}
],
"source": [
"import paddle"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "proprietary-disaster",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<function deepspeech.modules.repeat(xs: paddle.VarBase, *size: Any) -> paddle.VarBase>"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"paddle.Tensor.repeat"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "first-diagram",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<property at 0x7fb515eeeb88>"
]
},
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"paddle.Tensor.size"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "intelligent-david",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<function paddle.tensor.manipulation.concat(x, axis=0, name=None)>"
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"paddle.Tensor.cat"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "bronze-tenant",
"metadata": {},
"outputs": [],
"source": [
"a = paddle.to_tensor([12,32, 10, 12, 123,32 ,4])"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "balanced-bearing",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"7"
]
},
"execution_count": 13,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"a.size"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "extreme-republic",
"metadata": {},
"outputs": [],
"source": [
"def size(xs: paddle.Tensor, *args: int) -> paddle.Tensor:\n",
" nargs = len(args)\n",
" assert (nargs <= 1)\n",
" s = paddle.shape(xs)\n",
" if nargs == 1:\n",
" return s[args[0]]\n",
" else:\n",
" return s\n",
"\n",
"# logger.warn(\n",
"# \"override size of paddle.Tensor if exists or register, remove this when fixed!\"\n",
"# )\n",
"paddle.Tensor.size = size"
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "gross-addiction",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"Tensor(shape=[1], dtype=int32, place=CPUPlace, stop_gradient=True,\n",
" [7])"
]
},
"execution_count": 21,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"a.size(0)\n",
"a.size()"
]
},
{
"cell_type": "code",
"execution_count": 22,
"id": "adverse-dining",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"Tensor(shape=[1], dtype=int32, place=CPUPlace, stop_gradient=True,\n",
" [7])"
]
},
"execution_count": 22,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"a.size()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "popular-potato",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.0"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
此差异已折叠。
{
"cells": [
{
"cell_type": "code",
"execution_count": 32,
"id": "academic-surname",
"metadata": {},
"outputs": [],
"source": [
"import paddle\n",
"from paddle import nn"
]
},
{
"cell_type": "code",
"execution_count": 33,
"id": "fundamental-treasure",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Parameter containing:\n",
"Tensor(shape=[256], dtype=float32, place=CUDAPlace(0), stop_gradient=False,\n",
" [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.])\n",
"Parameter containing:\n",
"Tensor(shape=[256], dtype=float32, place=CUDAPlace(0), stop_gradient=False,\n",
" [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])\n"
]
}
],
"source": [
"L = nn.LayerNorm(256, epsilon=1e-12)\n",
"for p in L.parameters():\n",
" print(p)"
]
},
{
"cell_type": "code",
"execution_count": 34,
"id": "consolidated-elephant",
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n"
]
},
{
"cell_type": "code",
"execution_count": 46,
"id": "moderate-noise",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"float64\n"
]
}
],
"source": [
"x = np.random.randn(2, 51, 256)\n",
"print(x.dtype)"
]
},
{
"cell_type": "code",
"execution_count": 47,
"id": "cooked-progressive",
"metadata": {},
"outputs": [],
"source": [
"y = L(paddle.to_tensor(x, dtype='float32'))"
]
},
{
"cell_type": "code",
"execution_count": 48,
"id": "optimum-milwaukee",
"metadata": {},
"outputs": [],
"source": [
"import torch"
]
},
{
"cell_type": "code",
"execution_count": 49,
"id": "viral-indian",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Parameter containing:\n",
"tensor([1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,\n",
" 1., 1., 1., 1.], requires_grad=True)\n",
"Parameter containing:\n",
"tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
" 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
" requires_grad=True)\n"
]
}
],
"source": [
"TL = torch.nn.LayerNorm(256, eps=1e-12)\n",
"for p in TL.parameters():\n",
" print(p)"
]
},
{
"cell_type": "code",
"execution_count": 50,
"id": "skilled-vietnamese",
"metadata": {},
"outputs": [],
"source": [
"ty = TL(torch.tensor(x, dtype=torch.float32))"
]
},
{
"cell_type": "code",
"execution_count": 51,
"id": "incorrect-allah",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"False"
]
},
"execution_count": 51,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"np.allclose(y.numpy(), ty.detach().numpy())"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "prostate-cameroon",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 52,
"id": "governmental-surge",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"True"
]
},
"execution_count": 52,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"x = np.random.randn(2, 256)\n",
"y = L(paddle.to_tensor(x, dtype='float32'))\n",
"ty = TL(torch.tensor(x, dtype=torch.float32))\n",
"np.allclose(y.numpy(), ty.detach().numpy())"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "confidential-jacket",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.0"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
此差异已折叠。
{
"cells": [
{
"cell_type": "code",
"execution_count": 2,
"id": "designing-borough",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/workspace/DeepSpeech-2.x/tools/venv/lib/python3.7/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n",
" and should_run_async(code)\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"[[ 0.0000000e+00 0.0000000e+00 0.0000000e+00 ... 0.0000000e+00\n",
" 0.0000000e+00 0.0000000e+00]\n",
" [ 8.4147096e-01 8.0196178e-01 7.6172036e-01 ... 1.2409373e-04\n",
" 1.1547816e-04 1.0746076e-04]\n",
" [ 9.0929741e-01 9.5814437e-01 9.8704624e-01 ... 2.4818745e-04\n",
" 2.3095631e-04 2.1492151e-04]\n",
" ...\n",
" [ 3.7960774e-01 7.4510968e-01 7.3418564e-01 ... 1.2036801e-02\n",
" 1.1201146e-02 1.0423505e-02]\n",
" [-5.7338190e-01 -8.9752287e-02 -4.1488394e-02 ... 1.2160885e-02\n",
" 1.1316618e-02 1.0530960e-02]\n",
" [-9.9920684e-01 -8.5234123e-01 -7.8794664e-01 ... 1.2284970e-02\n",
" 1.1432089e-02 1.0638415e-02]]\n",
"True\n",
"True\n"
]
}
],
"source": [
"import torch\n",
"import math\n",
"import numpy as np\n",
"\n",
"max_len=100\n",
"d_model=256\n",
"\n",
"pe = torch.zeros(max_len, d_model)\n",
"position = torch.arange(0, max_len,\n",
" dtype=torch.float32).unsqueeze(1)\n",
"toruch_position = position\n",
"div_term = torch.exp(\n",
" torch.arange(0, d_model, 2, dtype=torch.float32) *\n",
" -(math.log(10000.0) / d_model))\n",
"tourch_div_term = div_term.cpu().detach().numpy()\n",
"\n",
"\n",
"\n",
"torhc_sin = torch.sin(position * div_term)\n",
"torhc_cos = torch.cos(position * div_term)\n",
"print(torhc_sin.cpu().detach().numpy())\n",
"np_sin = np.sin((position * div_term).cpu().detach().numpy())\n",
"np_cos = np.cos((position * div_term).cpu().detach().numpy())\n",
"print(np.allclose(np_sin, torhc_sin.cpu().detach().numpy()))\n",
"print(np.allclose(np_cos, torhc_cos.cpu().detach().numpy()))\n",
"pe[:, 0::2] = torhc_sin\n",
"pe[:, 1::2] = torhc_cos\n",
"tourch_pe = pe.cpu().detach().numpy()"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "swiss-referral",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"True\n",
"True\n",
"False\n",
"False\n",
"False\n",
"False\n",
"[[ 1. 1. 1. ... 1. 1.\n",
" 1. ]\n",
" [ 0.5403023 0.59737533 0.6479059 ... 1. 1.\n",
" 1. ]\n",
" [-0.41614684 -0.28628543 -0.1604359 ... 0.99999994 1.\n",
" 1. ]\n",
" ...\n",
" [-0.92514753 -0.66694194 -0.67894876 ... 0.9999276 0.99993724\n",
" 0.9999457 ]\n",
" [-0.81928825 -0.9959641 -0.999139 ... 0.99992603 0.999936\n",
" 0.99994457]\n",
" [ 0.03982088 -0.52298605 -0.6157435 ... 0.99992454 0.9999347\n",
" 0.99994344]]\n",
"----\n",
"[[ 1. 1. 1. ... 1. 1.\n",
" 1. ]\n",
" [ 0.54030234 0.59737533 0.6479059 ... 1. 1.\n",
" 1. ]\n",
" [-0.41614684 -0.28628543 -0.1604359 ... 1. 1.\n",
" 1. ]\n",
" ...\n",
" [-0.92514753 -0.66694194 -0.67894876 ... 0.9999276 0.9999373\n",
" 0.9999457 ]\n",
" [-0.81928825 -0.9959641 -0.999139 ... 0.99992603 0.999936\n",
" 0.99994457]\n",
" [ 0.03982088 -0.5229861 -0.6157435 ... 0.99992454 0.9999347\n",
" 0.99994344]]\n",
")))))))\n",
"[[ 0.0000000e+00 0.0000000e+00 0.0000000e+00 ... 0.0000000e+00\n",
" 0.0000000e+00 0.0000000e+00]\n",
" [ 8.4147096e-01 8.0196178e-01 7.6172036e-01 ... 1.2409373e-04\n",
" 1.1547816e-04 1.0746076e-04]\n",
" [ 9.0929741e-01 9.5814437e-01 9.8704624e-01 ... 2.4818745e-04\n",
" 2.3095631e-04 2.1492151e-04]\n",
" ...\n",
" [ 3.7960774e-01 7.4510968e-01 7.3418564e-01 ... 1.2036801e-02\n",
" 1.1201146e-02 1.0423505e-02]\n",
" [-5.7338190e-01 -8.9752287e-02 -4.1488394e-02 ... 1.2160885e-02\n",
" 1.1316618e-02 1.0530960e-02]\n",
" [-9.9920684e-01 -8.5234123e-01 -7.8794664e-01 ... 1.2284970e-02\n",
" 1.1432089e-02 1.0638415e-02]]\n",
"----\n",
"[[ 0.0000000e+00 0.0000000e+00 0.0000000e+00 ... 0.0000000e+00\n",
" 0.0000000e+00 0.0000000e+00]\n",
" [ 8.4147096e-01 8.0196178e-01 7.6172036e-01 ... 1.2409373e-04\n",
" 1.1547816e-04 1.0746076e-04]\n",
" [ 9.0929741e-01 9.5814437e-01 9.8704624e-01 ... 2.4818745e-04\n",
" 2.3095631e-04 2.1492151e-04]\n",
" ...\n",
" [ 3.7960774e-01 7.4510968e-01 7.3418564e-01 ... 1.2036801e-02\n",
" 1.1201146e-02 1.0423505e-02]\n",
" [-5.7338190e-01 -8.9752287e-02 -4.1488394e-02 ... 1.2160885e-02\n",
" 1.1316618e-02 1.0530960e-02]\n",
" [-9.9920684e-01 -8.5234123e-01 -7.8794664e-01 ... 1.2284970e-02\n",
" 1.1432089e-02 1.0638415e-02]]\n"
]
}
],
"source": [
"import paddle\n",
"paddle.set_device('cpu')\n",
"ppe = paddle.zeros((max_len, d_model), dtype='float32')\n",
"position = paddle.arange(0, max_len,\n",
" dtype='float32').unsqueeze(1)\n",
"print(np.allclose(position.numpy(), toruch_position))\n",
"div_term = paddle.exp(\n",
" paddle.arange(0, d_model, 2, dtype='float32') *\n",
" -(math.log(10000.0) / d_model))\n",
"print(np.allclose(div_term.numpy(), tourch_div_term))\n",
"\n",
"\n",
"\n",
"p_sin = paddle.sin(position * div_term)\n",
"p_cos = paddle.cos(position * div_term)\n",
"print(np.allclose(np_sin, p_sin.numpy(), rtol=1.e-6, atol=0))\n",
"print(np.allclose(np_cos, p_cos.numpy(), rtol=1.e-6, atol=0))\n",
"ppe[:, 0::2] = p_sin\n",
"ppe[:, 1::2] = p_cos\n",
"print(np.allclose(p_sin.numpy(), torhc_sin.cpu().detach().numpy()))\n",
"print(np.allclose(p_cos.numpy(), torhc_cos.cpu().detach().numpy()))\n",
"print(p_cos.numpy())\n",
"print(\"----\")\n",
"print(torhc_cos.cpu().detach().numpy())\n",
"print(\")))))))\")\n",
"print(p_sin.numpy())\n",
"print(\"----\")\n",
"print(torhc_sin.cpu().detach().numpy())"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "integrated-boards",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"False\n"
]
}
],
"source": [
"print(np.allclose(ppe.numpy(), pe.numpy()))"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "flying-reserve",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "revised-divide",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.7.0"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
[中文版](README_cn.md) # PaddlePaddle Speech to Any toolkit
# PaddlePaddle ASR toolkit
![License](https://img.shields.io/badge/license-Apache%202-red.svg) ![License](https://img.shields.io/badge/license-Apache%202-red.svg)
![python version](https://img.shields.io/badge/python-3.7+-orange.svg) ![python version](https://img.shields.io/badge/python-3.7+-orange.svg)
![support os](https://img.shields.io/badge/os-linux-yellow.svg) ![support os](https://img.shields.io/badge/os-linux-yellow.svg)
*PaddleASR* is an open-source implementation of end-to-end Automatic Speech Recognition (ASR) engine, with [PaddlePaddle](https://github.com/PaddlePaddle/Paddle) platform. Our vision is to empower both industrial application and academic research on speech recognition, via an easy-to-use, efficient, samller and scalable implementation, including training, inference & testing module, and deployment. *DeepSpeech* is an open-source implementation of end-to-end Automatic Speech Recognition engine, with [PaddlePaddle](https://github.com/PaddlePaddle/Paddle) platform. Our vision is to empower both industrial application and academic research on speech recognition, via an easy-to-use, efficient, samller and scalable implementation, including training, inference & testing module, and deployment.
## Features ## Features
See [feature list](doc/src/feature_list.md) for more information. See [feature list](docs/src/feature_list.md) for more information.
## Setup ## Setup
All tested under:
* Ubuntu 16.04
* python>=3.7 * python>=3.7
* paddlepaddle>=2.1.0 * paddlepaddle>=2.1.2
Please see [install](doc/src/install.md). Please see [install](docs/src/install.md).
## Getting Started ## Getting Started
Please see [Getting Started](doc/src/getting_started.md) and [tiny egs](examples/tiny/s0/README.md). Please see [Getting Started](docs/src/getting_started.md) and [tiny egs](examples/tiny/s0/README.md).
## More Information ## More Information
* [Data Prepration](doc/src/data_preparation.md) * [Data Prepration](docs/src/data_preparation.md)
* [Data Augmentation](doc/src/augmentation.md) * [Data Augmentation](docs/src/augmentation.md)
* [Ngram LM](doc/src/ngram_lm.md) * [Ngram LM](docs/src/ngram_lm.md)
* [Server Demo](doc/src/server.md) * [Benchmark](docs/src/benchmark.md)
* [Benchmark](doc/src/benchmark.md) * [Relased Model](docs/src/released_model.md)
* [Relased Model](doc/src/released_model.md)
* [FAQ](doc/src/faq.md)
## Questions and Help ## Questions and Help
...@@ -43,8 +41,8 @@ You are welcome to submit questions in [Github Discussions](https://github.com/P ...@@ -43,8 +41,8 @@ You are welcome to submit questions in [Github Discussions](https://github.com/P
## License ## License
DeepASR is provided under the [Apache-2.0 License](./LICENSE). DeepSpeech is provided under the [Apache-2.0 License](./LICENSE).
## Acknowledgement ## Acknowledgement
We depends on many open source repos. See [References](doc/src/reference.md) for more information. We depends on many open source repos. See [References](docs/src/reference.md) for more information.
[English](README.md)
# PaddlePaddle ASR toolkit
![License](https://img.shields.io/badge/license-Apache%202-red.svg)
![python version](https://img.shields.io/badge/python-3.7+-orange.svg)
![support os](https://img.shields.io/badge/os-linux-yellow.svg)
*PaddleASR*是一个采用[PaddlePaddle](https://github.com/PaddlePaddle/Paddle)平台的端到端自动语音识别(ASR)引擎的开源项目,
我们的愿景是为语音识别在工业应用和学术研究上,提供易于使用、高效、小型化和可扩展的工具,包括训练,推理,以及 部署。
## 特性
参看 [特性列表](doc/src/feature_list.md)
## 安装
* python>=3.7
* paddlepaddle>=2.1.0
参看 [安装](doc/src/install.md)
## 开始
请查看 [开始](doc/src/getting_started.md)[tiny egs](examples/tiny/s0/README.md)
## 更多信息
* [数据处理](doc/src/data_preparation.md)
* [数据增强](doc/src/augmentation.md)
* [语言模型](doc/src/ngram_lm.md)
* [服务部署](doc/src/server.md)
* [Benchmark](doc/src/benchmark.md)
* [Relased Model](doc/src/released_model.md)
* [FAQ](doc/src/faq.md)
## 问题和帮助
欢迎您在[Github讨论](https://github.com/PaddlePaddle/DeepSpeech/discussions)提交问题,[Github问题](https://github.com/PaddlePaddle/models/issues)中反馈bug。也欢迎您为这个项目做出贡献。
## License
DeepASR 遵循[Apache-2.0开源协议](./LICENSE)
## 感谢
开发中参考一些优秀的仓库,详情参见 [References](doc/src/reference.md)
# Benchmarks
## Acceleration with Multi-GPUs
We compare the training time with 1, 2, 4, 8 Tesla V100 GPUs (with a subset of LibriSpeech samples whose audio durations are between 6.0 and 7.0 seconds). And it shows that a **near-linear** acceleration with multiple GPUs has been achieved. In the following figure, the time (in seconds) cost for training is printed on the blue bars.
<img src="../images/multi_gpu_speedup.png" width=450>
| # of GPU | Acceleration Rate |
| -------- | --------------: |
| 1 | 1.00 X |
| 2 | 1.98 X |
| 4 | 3.73 X |
| 8 | 6.95 X |
`utils/profile.sh` provides such a demo profiling tool, you can change it as need.
# FAQ
1. 音频变速快慢到达什么晨读会影响识别率?
变速会提升识别效果,一般用0.9, 1.0, 1.1 的变速。
2. 音量大小到什么程度会影响识别率?
一般训练会固定音量到一个范围内,波动过大会影响训练,估计在10dB ~ 20dB吧。
3. 语音模型训练数据的最小时长要求时多少?
Aishell-1大约178h的数据,数据越多越好。
4. 那些噪声或背景生会影响识别率?
主要是人生干扰和低信噪比会影响识别率。
5. 单条语音数据的长度限制是多少?
一般训练的语音长度会限制在1s~6s之间,和训练配置有关。
6. 背景声在识别前是否需要分离出来,或做降噪处理?
需要分离的,需要结合具体场景考虑。
7. 模型是否带有VAD人生激活识别能力?
VAD是单独的模型或模块,模型不包含此能力。
8. 是否支持长语音识别?
一般过VAD后识别。
9. Mandarin LM Large语言模型需要的硬件配置时怎样的?
内存能放得下LM即可。
# Reference
* [wenet](https://github.com/mobvoi/wenet)
# Released Models
## Language Model Released
Language Model | Training Data | Token-based | Size | Descriptions
:-------------:| :------------:| :-----: | -----: | :-----------------
[English LM](https://deepspeech.bj.bcebos.com/en_lm/common_crawl_00.prune01111.trie.klm) | [CommonCrawl(en.00)](http://web-language-models.s3-website-us-east-1.amazonaws.com/ngrams/en/deduped/en.00.deduped.xz) | Word-based | 8.3 GB | Pruned with 0 1 1 1 1; <br/> About 1.85 billion n-grams; <br/> 'trie' binary with '-a 22 -q 8 -b 8'
[Mandarin LM Small](https://deepspeech.bj.bcebos.com/zh_lm/zh_giga.no_cna_cmn.prune01244.klm) | Baidu Internal Corpus | Char-based | 2.8 GB | Pruned with 0 1 2 4 4; <br/> About 0.13 billion n-grams; <br/> 'probing' binary with default settings
[Mandarin LM Large](https://deepspeech.bj.bcebos.com/zh_lm/zhidao_giga.klm) | Baidu Internal Corpus | Char-based | 70.4 GB | No Pruning; <br/> About 3.7 billion n-grams; <br/> 'probing' binary with default settings
此差异已折叠。
...@@ -21,7 +21,7 @@ To perform z-score normalization (zero-mean, unit stddev) upon audio features, w ...@@ -21,7 +21,7 @@ To perform z-score normalization (zero-mean, unit stddev) upon audio features, w
```bash ```bash
python3 utils/compute_mean_std.py \ python3 utils/compute_mean_std.py \
--num_samples 2000 \ --num_samples 2000 \
--specgram_type linear \ --spectrum_type linear \
--manifest_path examples/librispeech/data/manifest.train \ --manifest_path examples/librispeech/data/manifest.train \
--output_path examples/librispeech/data/mean_std.npz --output_path examples/librispeech/data/mean_std.npz
``` ```
......
此差异已折叠。
# Reference
We refer these repos to build `model` and `engine`:
* [delta](https://github.com/Delta-ML/delta.git)
* [espnet](https://github.com/espnet/espnet.git)
* [kaldi](https://github.com/kaldi-asr/kaldi.git)
* [wenet](https://github.com/mobvoi/wenet)
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
文件模式从 100644 更改为 100755
此差异已折叠。
此差异已折叠。
此差异已折叠。
文件模式从 100644 更改为 100755
此差异已折叠。
文件模式从 100644 更改为 100755
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
此差异已折叠。
文件模式从 100644 更改为 100755
此差异已折叠。
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册