DTLN (models: ailia-models)

.gitattributes

@@ -45,3 +45,4 @@ models/DTLN_pytorch[[:space:]](ONNX,[[:space:]]pth,[[:space:]]bin)/samples/audio
 models/DTLN_pytorch[[:space:]](ONNX,[[:space:]]pth,[[:space:]]bin)/samples/enhanced.wav filter=lfs diff=lfs merge=lfs -text
 models/DTLN_pytorch[[:space:]](ONNX,[[:space:]]pth,[[:space:]]bin)/samples/in.png filter=lfs diff=lfs merge=lfs -text
 models/DTLN_pytorch[[:space:]](ONNX,[[:space:]]pth,[[:space:]]bin)/samples/out.png filter=lfs diff=lfs merge=lfs -text
+models/ailia-models/code/1221-135766-0000.wav filter=lfs diff=lfs merge=lfs -text

models/ailia-models/code/1221-135766-0000.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29c67807e2be8a661b38ac883f3197be8d0550b3940eb3c09e24b72c0d0dde48
+size 397964

models/ailia-models/code/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2020 Nils L. Westhausen
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

models/ailia-models/code/README.md

@@ -0,0 +1,49 @@
+# Dual-signal Transformation LSTM Network
+
+## input
+
+audio file（16kHz)
+
+```
+LibriSpeech ASR corpus
+http://www.openslr.org/12
+1221-135766-0000.wav
+```
+
+## Output
+
+Audio file with noise removed
+
+## Usage
+Automatically downloads the onnx and prototxt files on the first run.
+It is necessary to be connected to the Internet while downloading.
+
+For the sample wav,
+```bash
+$ python3 dtln.py
+```
+
+If you want to specify the audio, put the file path after the `--input` option.
+```bash
+$ python3 dtln.py --input AUDIO_FILE
+```
+If you run by onnxruntime instead of ailia, you use --onnx option.
+
+
+
+## Reference
+
+- [Dual-signal Transformation LSTM Network](https://github.com/breizhn/DTLN)
+
+## Framework
+
+Pytorch
+
+## Model Format
+
+ONNX opset=11
+
+## Netron
+
+- [dtln1.onnx.prototxt](https://netron.app/?url=https://storage.googleapis.com/ailia-models/dtln/dtln1.onnx.prototxt)  
+- [dtln2.onnx.prototxt](https://netron.app/?url=https://storage.googleapis.com/ailia-models/dtln2/dtln2.onnx.prototxt)

models/ailia-models/code/dtln.py

@@ -0,0 +1,213 @@
+import sys
+import time
+from logging import getLogger
+import onnxruntime
+
+import numpy as np
+import soundfile as sf
+
+import ailia
+
+# import original modules
+sys.path.append('../../util')
+from model_utils import check_and_download_models  # noqa
+from arg_utils import get_base_parser, get_savepath, update_parser  # noqa
+
+logger = getLogger(__name__)
+
+# ======================
+# Parameters
+# ======================
+
+WEIGHT1_PATH = "dtln1.onnx"
+MODEL1_PATH = "dtln1.onnx.prototxt"
+WEIGHT2_PATH = "dtln2.onnx"
+MODEL2_PATH = "dtln2.onnx.prototxt"
+REMOTE_PATH = 'https://storage.googleapis.com/ailia-models/dtln/'
+
+SAMPLE_RATE = 16000
+
+WAV_PATH = '1221-135766-0000.wav'
+SAVE_WAV_PATH = 'output.wav'
+
+# ======================
+# Arguemnt Parser Config
+# ======================
+
+parser = get_base_parser(
+    'Dual-signal Transformation LSTM Network', WAV_PATH, SAVE_WAV_PATH, input_ftype='audio'
+)
+parser.add_argument(
+    '--onnx',
+    action='store_true',
+    help='By default, the ailia SDK is used, but with this option, you can switch to using ONNX Runtime'
+)
+parser.add_argument(
+    '--shift',
+    default=128, type=int,
+)
+args = update_parser(parser)
+
+block_shift = args.shift
+
+# ======================
+# Main functions
+# ======================
+
+def predict(audio,models):
+    block_len = 512
+    out_file = np.zeros((len(audio)))
+    # create buffer
+    in_buffer = np.zeros((block_len)).astype('float32')
+    out_buffer = np.zeros((block_len)).astype('float32')
+    # calculate number of blocks
+    num_blocks = (audio.shape[0] - (block_len-block_shift)) // block_shift
+    # iterate over the number of blcoks  
+    time_array = []      
+
+    inp_shape = [(1, 1, 257) ,(1, 2, 128, 2)]
+    model_input_names_1 = ['input_2', 'input_3']
+    interpreter_1 = models[0]
+    interpreter_2 = models[1]
+    model_inputs_1 = {}
+    model_inputs_1[model_input_names_1[0]] = np.zeros(
+                    [dim if isinstance(dim, int) else 1 for dim in inp_shape[0]],
+                    dtype=np.float32)
+    model_inputs_1[model_input_names_1[1]] = np.zeros(
+                    [dim if isinstance(dim, int) else 1 for dim in inp_shape[1]],
+                    dtype=np.float32)
+
+    model_input_names_2 = ['input_4','input_5']
+    model_inputs_2 = {}
+    inp_shape = [(1, 1, 512) ,(1, 2, 128, 2)]
+    model_inputs_2[model_input_names_2[0]] = np.zeros(
+                [dim if isinstance(dim, int) else 1 for dim in inp_shape[0]],
+                dtype=np.float32)
+    model_inputs_2[model_input_names_2[1]] = np.zeros(
+                    [dim if isinstance(dim, int) else 1 for dim in inp_shape[1]],
+                    dtype=np.float32)
+
+
+    for idx in range(num_blocks):
+        start_time = time.time()
+        # shift values and write to buffer
+        in_buffer[:-block_shift] = in_buffer[block_shift:]
+        in_buffer[-block_shift:] = audio[idx*block_shift:(idx*block_shift)+block_shift]
+        # calculate fft of input block
+        in_block_fft = np.fft.rfft(in_buffer)
+        in_mag = np.abs(in_block_fft)
+        in_phase = np.angle(in_block_fft)
+        # reshape magnitude to input dimensions
+        in_mag = np.reshape(in_mag, (1,1,-1)).astype('float32')
+        # set block to input
+        model_inputs_1[model_input_names_1[0]] = in_mag
+        # run calculation 
+        if args.onnx:
+            model_outputs_1 = interpreter_1.run([],{'input_2':model_inputs_1['input_2'],'input_3':model_inputs_1['input_3']})
+        else:
+            inputs = [model_inputs_1['input_2'],model_inputs_1['input_3']]
+            model_outputs_1 = interpreter_1.run(inputs)
+        # get the output of the first block
+        out_mask = model_outputs_1[0]
+        # set out states back to input
+        model_inputs_1[model_input_names_1[1]] = model_outputs_1[1]  
+        # calculate the ifft
+        estimated_complex = in_mag * out_mask * np.exp(1j * in_phase)
+        estimated_block = np.fft.irfft(estimated_complex)
+        # reshape the time domain block
+        estimated_block = np.reshape(estimated_block, (1,1,-1)).astype('float32')
+        # set tensors to the second block
+        model_inputs_2[model_input_names_2[0]] = estimated_block
+        # run calculation
+        if args.onnx:
+            model_outputs_2 = interpreter_2.run([],{'input_4':model_inputs_2['input_4'],'input_5':model_inputs_2['input_5']})
+        else:
+            inputs = [model_inputs_2['input_4'],model_inputs_2['input_5']]
+            model_outputs_2 = interpreter_2.run(inputs)
+        # get output
+        out_block = model_outputs_2[0]
+        # set out states back to input
+        model_inputs_2[model_input_names_2[1]] = model_outputs_2[1]
+        # shift values and write to buffer
+        out_buffer[:-block_shift] = out_buffer[block_shift:]
+        out_buffer[-block_shift:] = np.zeros((block_shift))
+        out_buffer  += np.squeeze(out_block)
+        # write block to output file
+        out_file[idx*block_shift:(idx*block_shift)+block_shift] = out_buffer[:block_shift]
+        time_array.append(time.time()-start_time)
+    return out_file
+
+def recognize_from_audio(models):
+
+    inp_shape = [(1, 1, 257) ,(1, 2, 128, 2)]
+    model_input_names_1 = ['input_2', 'input_3']
+
+    model_inputs_1 = {}
+    model_inputs_1[model_input_names_1[0]] = np.zeros(
+                    [dim if isinstance(dim, int) else 1 for dim in inp_shape[0]],
+                    dtype=np.float32)
+    model_inputs_1[model_input_names_1[1]] = np.zeros(
+                    [dim if isinstance(dim, int) else 1 for dim in inp_shape[1]],
+                    dtype=np.float32)
+
+    model_input_names_2 = ['input_4','input_5']
+    model_inputs_2 = {}
+    inp_shape = [(1, 1, 512) ,(1, 2, 128, 2)]
+    model_inputs_2[model_input_names_2[0]] = np.zeros(
+                [dim if isinstance(dim, int) else 1 for dim in inp_shape[0]],
+                dtype=np.float32)
+    model_inputs_2[model_input_names_2[1]] = np.zeros(
+                    [dim if isinstance(dim, int) else 1 for dim in inp_shape[1]],
+                    dtype=np.float32)
+
+
+    # input audio loop
+    for audio_path in args.input:
+        logger.info(audio_path)
+
+        # load audio file
+        audio,fs = sf.read(audio_path)
+        # check for sampling rate
+        if fs != 16000:
+            print('This model only supports 16k sampling rate.')
+            continue
+
+        # inference
+        logger.info('Start inference...')
+        if args.benchmark:
+            logger.info('BENCHMARK mode')
+            start = int(round(time.time() * 1000))
+            output,sr = predict(audio, models)
+            end = int(round(time.time() * 1000))
+            estimation_time = (end - start)
+            logger.info(f'\ttotal processing time {estimation_time} ms')
+        else:
+            output = predict(audio, models)
+
+        # save result
+        savepath = get_savepath(args.savepath, audio_path, ext='.wav')
+        logger.info(f'saved at : {savepath}')
+        sf.write(savepath, output, fs)
+
+    logger.info('Script finished successfully.')
+
+
+def main():
+    check_and_download_models(WEIGHT1_PATH, MODEL1_PATH, REMOTE_PATH)
+    check_and_download_models(WEIGHT2_PATH, MODEL2_PATH, REMOTE_PATH)
+
+    env_id = args.env_id
+
+    if args.onnx:
+        models = [onnxruntime.InferenceSession(WEIGHT1_PATH),
+                  onnxruntime.InferenceSession(WEIGHT2_PATH)]
+    else:
+        models = [ailia.Net(MODEL1_PATH,WEIGHT1_PATH, env_id = env_id),
+                  ailia.Net(MODEL2_PATH,WEIGHT2_PATH, env_id = env_id)]
+
+    # initialize
+    recognize_from_audio(models)
+
+
+if __name__ == '__main__':
+    main()

models/ailia-models/dtln1.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22b91cae3855e5a0620e66a917ca6c82c58db0e842c770f58d86751c5e8d4ae3
+size 1458237

models/ailia-models/dtln1.onnx.prototxt

@@ -0,0 +1,705 @@
+ir_version: 6
+producer_name: "OnnxMLTools"
+producer_version: "1.7.0"
+model_version: 0
+graph {
+  name: "model_1"
+  node {
+    input: "input_2"
+    output: "lstm_4_X"
+    name: "Transpose2"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 1
+      ints: 0
+      ints: 2
+    }
+    doc_string: ""
+  }
+  node {
+    input: "input_3"
+    input: "input_3:01_cropping_start"
+    input: "input_3:01_cropping_end"
+    input: "input_3:01_cropping_axes"
+    input: "input_3:01_cropping_steps"
+    output: "input_3:01_cropping0"
+    name: "input_3:01_cropping"
+    op_type: "Slice"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "input_3"
+    input: "input_3:01_cropping_start1"
+    input: "input_3:01_cropping_end1"
+    input: "input_3:01_cropping_axes1"
+    input: "input_3:01_cropping_steps1"
+    output: "input_3:01_cropping01"
+    name: "input_3:01_cropping_1"
+    op_type: "Slice"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "input_3"
+    input: "input_3:01_cropping_start2"
+    input: "input_3:01_cropping_end2"
+    input: "input_3:01_cropping_axes2"
+    input: "input_3:01_cropping_steps2"
+    output: "input_3:01_cropping02"
+    name: "input_3:01_cropping_2"
+    op_type: "Slice"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "input_3"
+    input: "input_3:01_cropping_start3"
+    input: "input_3:01_cropping_end3"
+    input: "input_3:01_cropping_axes3"
+    input: "input_3:01_cropping_steps3"
+    output: "input_3:01_cropping03"
+    name: "input_3:01_cropping_3"
+    op_type: "Slice"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "input_3:01_cropping0"
+    output: "strided_slice_3:0"
+    name: "input_3:01_squeeze"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+      ints: 3
+    }
+  }
+  node {
+    input: "input_3:01_cropping01"
+    output: "strided_slice_2:0"
+    name: "input_3:01_squeeze_1"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+      ints: 3
+    }
+  }
+  node {
+    input: "input_3:01_cropping02"
+    output: "strided_slice_1:0"
+    name: "input_3:01_squeeze_2"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+      ints: 3
+    }
+  }
+  node {
+    input: "input_3:01_cropping03"
+    output: "strided_slice:0"
+    name: "input_3:01_squeeze_3"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+      ints: 3
+    }
+  }
+  node {
+    input: "strided_slice_2:0"
+    output: "lstm_5_initial_h"
+    name: "Unsqueeze"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "strided_slice_3:0"
+    output: "lstm_5_initial_c"
+    name: "Unsqueeze1"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "strided_slice:0"
+    output: "lstm_4_initial_h"
+    name: "Unsqueeze2"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "strided_slice_1:0"
+    output: "lstm_4_initial_c"
+    name: "Unsqueeze3"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "lstm_4_X"
+    input: "lstm_4_W"
+    input: "lstm_4_R"
+    input: "lstm_4_B"
+    input: ""
+    input: "lstm_4_initial_h"
+    input: "lstm_4_initial_c"
+    input: ""
+    output: "lstm_4_Y"
+    output: "lstm_4_Y_h"
+    output: "lstm_4_Y_c"
+    name: "lstm_4"
+    op_type: "LSTM"
+    domain: ""
+    attribute {
+      name: "activations"
+      type: STRINGS
+      strings: "Sigmoid"
+      strings: "Tanh"
+      strings: "Tanh"
+    }
+    attribute {
+      name: "direction"
+      type: STRING
+      s: "forward"
+    }
+    attribute {
+      name: "hidden_size"
+      type: INT
+      i: 128
+    }
+  }
+  node {
+    input: "lstm_4_Y"
+    output: "lstm_4_y_transposed"
+    name: "Transpose3"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 2
+      ints: 1
+      ints: 0
+      ints: 3
+    }
+  }
+  node {
+    input: "lstm_4_Y_h"
+    output: "lstm_4/Identity_1:0"
+    name: "Squeeze4"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "lstm_4_Y_c"
+    output: "lstm_4/Identity_2:0"
+    name: "Squeeze5"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "lstm_4/Identity_1:0"
+    output: "stack_unsqueeze00"
+    name: "stack_unsqueeze0"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "lstm_4/Identity_2:0"
+    output: "stack_1_unsqueeze00"
+    name: "stack_1_unsqueeze0"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "lstm_4_y_transposed"
+    output: "lstm_4/Identity:0"
+    name: "Squeeze3"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "lstm_4/Identity:0"
+    output: "lstm_5_X"
+    name: "Transpose"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 1
+      ints: 0
+      ints: 2
+    }
+    doc_string: ""
+  }
+  node {
+    input: "lstm_5_X"
+    input: "lstm_5_W"
+    input: "lstm_5_R"
+    input: "lstm_5_B"
+    input: ""
+    input: "lstm_5_initial_h"
+    input: "lstm_5_initial_c"
+    input: ""
+    output: "lstm_5_Y"
+    output: "lstm_5_Y_h"
+    output: "lstm_5_Y_c"
+    name: "lstm_5"
+    op_type: "LSTM"
+    domain: ""
+    attribute {
+      name: "activations"
+      type: STRINGS
+      strings: "Sigmoid"
+      strings: "Tanh"
+      strings: "Tanh"
+    }
+    attribute {
+      name: "direction"
+      type: STRING
+      s: "forward"
+    }
+    attribute {
+      name: "hidden_size"
+      type: INT
+      i: 128
+    }
+  }
+  node {
+    input: "lstm_5_Y"
+    output: "lstm_5_y_transposed"
+    name: "Transpose1"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 2
+      ints: 1
+      ints: 0
+      ints: 3
+    }
+  }
+  node {
+    input: "lstm_5_Y_h"
+    output: "lstm_5/Identity_1:0"
+    name: "Squeeze1"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "lstm_5_Y_c"
+    output: "lstm_5/Identity_2:0"
+    name: "Squeeze2"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "lstm_5/Identity_1:0"
+    output: "stack_unsqueeze10"
+    name: "stack_unsqueeze1"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "lstm_5/Identity_2:0"
+    output: "stack_1_unsqueeze10"
+    name: "stack_1_unsqueeze1"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "lstm_5_y_transposed"
+    output: "lstm_5/Identity:0"
+    name: "Squeeze"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "lstm_5/Identity:0"
+    input: "dense_2/kernel:0"
+    output: "dense_20"
+    name: "dense_2"
+    op_type: "MatMul"
+    domain: ""
+  }
+  node {
+    input: "stack_unsqueeze00"
+    input: "stack_unsqueeze10"
+    output: "stack:0"
+    name: "stack_concat"
+    op_type: "Concat"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 1
+    }
+  }
+  node {
+    input: "stack_1_unsqueeze00"
+    input: "stack_1_unsqueeze10"
+    output: "stack_1:0"
+    name: "stack_1_concat"
+    op_type: "Concat"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 1
+    }
+  }
+  node {
+    input: "stack:0"
+    output: "stack_2_unsqueeze00"
+    name: "stack_2_unsqueeze0"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: -1
+    }
+  }
+  node {
+    input: "stack_1:0"
+    output: "stack_2_unsqueeze10"
+    name: "stack_2_unsqueeze1"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: -1
+    }
+  }
+  node {
+    input: "dense_20"
+    input: "dense_2/bias:0"
+    output: "biased_tensor_name"
+    name: "Add"
+    op_type: "Add"
+    domain: ""
+  }
+  node {
+    input: "biased_tensor_name"
+    output: "activation_2"
+    name: "Sigmoid"
+    op_type: "Sigmoid"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "stack_2_unsqueeze00"
+    input: "stack_2_unsqueeze10"
+    output: "tf_op_layer_stack_2"
+    name: "stack_2_concat"
+    op_type: "Concat"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: -1
+    }
+    doc_string: ""
+  }
+  initializer {
+      dims: 128
+      dims: 257
+      data_type: 1
+      name: "dense_2/kernel:0"
+  }
+  initializer {
+      dims: 257
+      data_type: 1
+      name: "dense_2/bias:0"
+  }
+  initializer {
+      dims: 1
+      dims: 512
+      dims: 128
+      data_type: 1
+      name: "lstm_5_W"
+  }
+  initializer {
+      dims: 1
+      dims: 512
+      dims: 128
+      data_type: 1
+      name: "lstm_5_R"
+  }
+  initializer {
+      dims: 1
+      dims: 1024
+      data_type: 1
+      name: "lstm_5_B"
+  }
+  initializer {
+      dims: 1
+      dims: 512
+      dims: 257
+      data_type: 1
+      name: "lstm_4_W"
+  }
+  initializer {
+      dims: 1
+      dims: 512
+      dims: 128
+      data_type: 1
+      name: "lstm_4_R"
+  }
+  initializer {
+      dims: 1
+      dims: 1024
+      data_type: 1
+      name: "lstm_4_B"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_start"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_end"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_axes"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_steps"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_start1"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_end1"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_axes1"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_steps1"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_start2"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_end2"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_axes2"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_steps2"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_start3"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_end3"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_axes3"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_3:01_cropping_steps3"
+  }
+  input {
+    name: "input_2"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 257
+          }
+        }
+      }
+    }
+  }
+  input {
+    name: "input_3"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 128
+          }
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "activation_2"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 257
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "tf_op_layer_stack_2"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 128
+          }
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+}
+opset_import {
+  version: 11
+}

models/ailia-models/dtln2.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e20c92f9233fccf29cddf86970d0d0161a03aebccc26d6f4d5639c4d5ec2e639
+size 2510010

models/ailia-models/dtln2.onnx.prototxt

@@ -0,0 +1,1123 @@
+ir_version: 6
+producer_name: "OnnxMLTools"
+producer_version: "1.7.0"
+model_version: 0
+graph {
+  name: "model_2"
+  node {
+    input: "input_4"
+    output: "adjusted_input1"
+    name: "Transpose2"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 0
+      ints: 2
+      ints: 1
+    }
+    doc_string: ""
+  }
+  node {
+    input: "input_5"
+    input: "input_5:01_cropping_start"
+    input: "input_5:01_cropping_end"
+    input: "input_5:01_cropping_axes"
+    input: "input_5:01_cropping_steps"
+    output: "input_5:01_cropping0"
+    name: "input_5:01_cropping"
+    op_type: "Slice"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "input_5"
+    input: "input_5:01_cropping_start1"
+    input: "input_5:01_cropping_end1"
+    input: "input_5:01_cropping_axes1"
+    input: "input_5:01_cropping_steps1"
+    output: "input_5:01_cropping01"
+    name: "input_5:01_cropping_1"
+    op_type: "Slice"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "input_5"
+    input: "input_5:01_cropping_start2"
+    input: "input_5:01_cropping_end2"
+    input: "input_5:01_cropping_axes2"
+    input: "input_5:01_cropping_steps2"
+    output: "input_5:01_cropping02"
+    name: "input_5:01_cropping_2"
+    op_type: "Slice"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "input_5"
+    input: "input_5:01_cropping_start3"
+    input: "input_5:01_cropping_end3"
+    input: "input_5:01_cropping_axes3"
+    input: "input_5:01_cropping_steps3"
+    output: "input_5:01_cropping03"
+    name: "input_5:01_cropping_3"
+    op_type: "Slice"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "adjusted_input1"
+    input: "conv1d_2/kernel:0"
+    output: "convolution_output1"
+    name: "conv1d_2"
+    op_type: "Conv"
+    domain: ""
+    attribute {
+      name: "auto_pad"
+      type: STRING
+      s: "VALID"
+    }
+    attribute {
+      name: "dilations"
+      type: INTS
+      ints: 1
+    }
+    attribute {
+      name: "group"
+      type: INT
+      i: 1
+    }
+    attribute {
+      name: "kernel_shape"
+      type: INTS
+      ints: 1
+    }
+    attribute {
+      name: "strides"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "input_5:01_cropping0"
+    output: "model_2/tf_op_layer_strided_slice_5/strided_slice_5:0"
+    name: "input_5:01_squeeze"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+      ints: 3
+    }
+  }
+  node {
+    input: "input_5:01_cropping01"
+    output: "model_2/tf_op_layer_strided_slice_7/strided_slice_7:0"
+    name: "input_5:01_squeeze_1"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+      ints: 3
+    }
+  }
+  node {
+    input: "input_5:01_cropping02"
+    output: "model_2/tf_op_layer_strided_slice_4/strided_slice_4:0"
+    name: "input_5:01_squeeze_2"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+      ints: 3
+    }
+  }
+  node {
+    input: "input_5:01_cropping03"
+    output: "model_2/tf_op_layer_strided_slice_6/strided_slice_6:0"
+    name: "input_5:01_squeeze_3"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+      ints: 3
+    }
+  }
+  node {
+    input: "convolution_output1"
+    output: "transpose_output1"
+    name: "Transpose3"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 0
+      ints: 2
+      ints: 1
+    }
+  }
+  node {
+    input: "model_2/tf_op_layer_strided_slice_4/strided_slice_4:0"
+    input: "model_2/lstm_6/MatMul_1/ReadVariableOp/resource:0"
+    output: "model_2/lstm_6/MatMul_1:0"
+    name: "model_2/lstm_6/MatMul_1_add"
+    op_type: "MatMul"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/tf_op_layer_strided_slice_6/strided_slice_6:0"
+    input: "model_2/lstm_7/MatMul_1/ReadVariableOp/resource:0"
+    output: "model_2/lstm_7/MatMul_1:0"
+    name: "model_2/lstm_7/MatMul_1_add"
+    op_type: "MatMul"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "transpose_output1"
+    output: "model_2/instant_layer_normalization_1/Mean:0"
+    name: "model_2/instant_layer_normalization_1/Mean_reduce_min"
+    op_type: "ReduceMean"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: -1
+    }
+    attribute {
+      name: "keepdims"
+      type: INT
+      i: 1
+    }
+    doc_string: ""
+  }
+  node {
+    input: "transpose_output1"
+    input: "model_2/instant_layer_normalization_1/Mean:0"
+    output: "model_2/instant_layer_normalization_1/sub_1:0"
+    name: "model_2/instant_layer_normalization_1/sub_1"
+    op_type: "Sub"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "transpose_output1"
+    input: "model_2/instant_layer_normalization_1/Mean:0"
+    output: "model_2/instant_layer_normalization_1/sub:0"
+    name: "model_2/instant_layer_normalization_1/sub"
+    op_type: "Sub"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/instant_layer_normalization_1/sub:0"
+    input: "model_2/instant_layer_normalization_1/sub:0"
+    output: "model_2/instant_layer_normalization_1/Square:0"
+    name: "model_2/instant_layer_normalization_1/Square"
+    op_type: "Mul"
+    domain: ""
+  }
+  node {
+    input: "model_2/instant_layer_normalization_1/Square:0"
+    output: "model_2/instant_layer_normalization_1/Mean_1:0"
+    name: "model_2/instant_layer_normalization_1/Mean_1_reduce_min"
+    op_type: "ReduceMean"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: -1
+    }
+    attribute {
+      name: "keepdims"
+      type: INT
+      i: 1
+    }
+  }
+  node {
+    input: "model_2/instant_layer_normalization_1/Mean_1:0"
+    input: "model_2/instant_layer_normalization_1/add/y:0"
+    output: "model_2/instant_layer_normalization_1/add:0"
+    name: "model_2/instant_layer_normalization_1/add"
+    op_type: "Add"
+    domain: ""
+  }
+  node {
+    input: "model_2/instant_layer_normalization_1/add:0"
+    output: "model_2/instant_layer_normalization_1/Sqrt:0"
+    name: "model_2/instant_layer_normalization_1/Sqrt"
+    op_type: "Sqrt"
+    domain: ""
+  }
+  node {
+    input: "model_2/instant_layer_normalization_1/sub_1:0"
+    input: "model_2/instant_layer_normalization_1/Sqrt:0"
+    output: "model_2/instant_layer_normalization_1/truediv:0"
+    name: "model_2/instant_layer_normalization_1/truediv"
+    op_type: "Div"
+    domain: ""
+  }
+  node {
+    input: "model_2/instant_layer_normalization_1/truediv:0"
+    input: "model_2/instant_layer_normalization_1/mul/ReadVariableOp/resource:0"
+    output: "model_2/instant_layer_normalization_1/mul:0"
+    name: "model_2/instant_layer_normalization_1/mul"
+    op_type: "Mul"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/instant_layer_normalization_1/mul:0"
+    input: "model_2/instant_layer_normalization_1/add_1/ReadVariableOp/resource:0"
+    output: "model_2/instant_layer_normalization_1/add_1:0"
+    name: "model_2/instant_layer_normalization_1/add_1"
+    op_type: "Add"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/instant_layer_normalization_1/add_1:0"
+    output: "model_2/lstm_6/transpose:0"
+    name: "model_2/lstm_6/transpose"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 1
+      ints: 0
+      ints: 2
+    }
+  }
+  node {
+    input: "model_2/lstm_6/transpose:0"
+    output: "model_2/lstm_6/unstack_split0"
+    name: "model_2/lstm_6/unstack_split"
+    op_type: "Split"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 0
+    }
+  }
+  node {
+    input: "model_2/lstm_6/unstack_split0"
+    output: "model_2/lstm_6/unstack:0"
+    name: "model_2/lstm_6/unstack_squeeze_0"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "model_2/lstm_6/unstack:0"
+    input: "model_2/lstm_6/MatMul/ReadVariableOp/resource:0"
+    output: "model_2/lstm_6/MatMul:0"
+    name: "model_2/lstm_6/MatMul_add"
+    op_type: "MatMul"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/lstm_6/MatMul:0"
+    input: "model_2/lstm_6/MatMul_1:0"
+    output: "model_2/lstm_6/add:0"
+    name: "model_2/lstm_6/add"
+    op_type: "Add"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/add:0"
+    input: "model_2/lstm_6/BiasAdd/ReadVariableOp/resource:0"
+    output: "model_2/lstm_6/BiasAdd:0"
+    name: "model_2/lstm_6/BiasAdd_add"
+    op_type: "Add"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/lstm_6/BiasAdd:0"
+    output: "model_2/lstm_6/split:0"
+    output: "model_2/lstm_6/split:1"
+    output: "model_2/lstm_6/split:2"
+    output: "model_2/lstm_6/split:3"
+    name: "model_2/lstm_6/split/split_dim:0_split"
+    op_type: "Split"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 1
+    }
+  }
+  node {
+    input: "model_2/lstm_6/split:3"
+    output: "model_2/lstm_6/Sigmoid_2:0"
+    name: "model_2/lstm_6/Sigmoid_2"
+    op_type: "Sigmoid"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/split:1"
+    output: "model_2/lstm_6/Sigmoid_1:0"
+    name: "model_2/lstm_6/Sigmoid_1"
+    op_type: "Sigmoid"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/split:0"
+    output: "model_2/lstm_6/Sigmoid:0"
+    name: "model_2/lstm_6/Sigmoid"
+    op_type: "Sigmoid"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/split:2"
+    output: "model_2/lstm_6/Tanh:0"
+    name: "model_2/lstm_6/Tanh"
+    op_type: "Tanh"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/Sigmoid_1:0"
+    input: "model_2/tf_op_layer_strided_slice_5/strided_slice_5:0"
+    output: "model_2/lstm_6/mul:0"
+    name: "model_2/lstm_6/mul"
+    op_type: "Mul"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/Sigmoid:0"
+    input: "model_2/lstm_6/Tanh:0"
+    output: "model_2/lstm_6/mul_1:0"
+    name: "model_2/lstm_6/mul_1"
+    op_type: "Mul"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/mul:0"
+    input: "model_2/lstm_6/mul_1:0"
+    output: "model_2/lstm_6/add_1:0"
+    name: "model_2/lstm_6/add_1"
+    op_type: "Add"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/add_1:0"
+    output: "model_2/tf_op_layer_stack_4/stack_4_unsqueeze00"
+    name: "model_2/tf_op_layer_stack_4/stack_4_unsqueeze0"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "model_2/lstm_6/add_1:0"
+    output: "model_2/lstm_6/Tanh_1:0"
+    name: "model_2/lstm_6/Tanh_1"
+    op_type: "Tanh"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/Sigmoid_2:0"
+    input: "model_2/lstm_6/Tanh_1:0"
+    output: "model_2/lstm_6/mul_2:0"
+    name: "model_2/lstm_6/mul_2"
+    op_type: "Mul"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_6/mul_2:0"
+    output: "model_2/tf_op_layer_stack_3/stack_3_unsqueeze00"
+    name: "model_2/tf_op_layer_stack_3/stack_3_unsqueeze0"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "model_2/lstm_6/mul_2:0"
+    output: "model_2/lstm_6/stack_unsqueeze00"
+    name: "model_2/lstm_6/stack_unsqueeze0"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "model_2/lstm_6/stack_unsqueeze00"
+    output: "model_2/lstm_6/stack:0"
+    name: "model_2/lstm_6/stack_concat"
+    op_type: "Concat"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 0
+    }
+  }
+  node {
+    input: "model_2/lstm_6/stack:0"
+    output: "model_2/lstm_7/unstack_split0"
+    name: "model_2/lstm_7/unstack_split"
+    op_type: "Split"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 0
+    }
+    doc_string: ""
+  }
+  node {
+    input: "model_2/lstm_7/unstack_split0"
+    output: "model_2/lstm_7/unstack:0"
+    name: "model_2/lstm_7/unstack_squeeze_0"
+    op_type: "Squeeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "model_2/lstm_7/unstack:0"
+    input: "model_2/lstm_7/MatMul/ReadVariableOp/resource:0"
+    output: "model_2/lstm_7/MatMul:0"
+    name: "model_2/lstm_7/MatMul_add"
+    op_type: "MatMul"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/lstm_7/MatMul:0"
+    input: "model_2/lstm_7/MatMul_1:0"
+    output: "model_2/lstm_7/add:0"
+    name: "model_2/lstm_7/add"
+    op_type: "Add"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_7/add:0"
+    input: "model_2/lstm_7/BiasAdd/ReadVariableOp/resource:0"
+    output: "model_2/lstm_7/BiasAdd:0"
+    name: "model_2/lstm_7/BiasAdd_add"
+    op_type: "Add"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/lstm_7/BiasAdd:0"
+    output: "model_2/lstm_7/split:0"
+    output: "model_2/lstm_7/split:1"
+    output: "model_2/lstm_7/split:2"
+    output: "model_2/lstm_7/split:3"
+    name: "model_2/lstm_7/split/split_dim:0_split"
+    op_type: "Split"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 1
+    }
+  }
+  node {
+    input: "model_2/lstm_7/split:3"
+    output: "model_2/lstm_7/Sigmoid_2:0"
+    name: "model_2/lstm_7/Sigmoid_2"
+    op_type: "Sigmoid"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_7/split:1"
+    output: "model_2/lstm_7/Sigmoid_1:0"
+    name: "model_2/lstm_7/Sigmoid_1"
+    op_type: "Sigmoid"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_7/split:0"
+    output: "model_2/lstm_7/Sigmoid:0"
+    name: "model_2/lstm_7/Sigmoid"
+    op_type: "Sigmoid"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_7/split:2"
+    output: "model_2/lstm_7/Tanh:0"
+    name: "model_2/lstm_7/Tanh"
+    op_type: "Tanh"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_7/Sigmoid_1:0"
+    input: "model_2/tf_op_layer_strided_slice_7/strided_slice_7:0"
+    output: "model_2/lstm_7/mul:0"
+    name: "model_2/lstm_7/mul"
+    op_type: "Mul"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_7/Sigmoid:0"
+    input: "model_2/lstm_7/Tanh:0"
+    output: "model_2/lstm_7/mul_1:0"
+    name: "model_2/lstm_7/mul_1"
+    op_type: "Mul"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_7/mul:0"
+    input: "model_2/lstm_7/mul_1:0"
+    output: "model_2/lstm_7/add_1:0"
+    name: "model_2/lstm_7/add_1"
+    op_type: "Add"
+    domain: ""
+  }
+  node {
+    input: "model_2/lstm_7/add_1:0"
+    output: "model_2/tf_op_layer_stack_4/stack_4_unsqueeze10"
+    name: "model_2/tf_op_layer_stack_4/stack_4_unsqueeze1"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "model_2/lstm_7/add_1:0"
+    output: "model_2/lstm_7/Tanh_1:0"
+    name: "model_2/lstm_7/Tanh_1"
+    op_type: "Tanh"
+    domain: ""
+  }
+  node {
+    input: "model_2/tf_op_layer_stack_4/stack_4_unsqueeze00"
+    input: "model_2/tf_op_layer_stack_4/stack_4_unsqueeze10"
+    output: "model_2/tf_op_layer_stack_4/stack_4:0"
+    name: "model_2/tf_op_layer_stack_4/stack_4_concat"
+    op_type: "Concat"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 1
+    }
+  }
+  node {
+    input: "model_2/lstm_7/Sigmoid_2:0"
+    input: "model_2/lstm_7/Tanh_1:0"
+    output: "model_2/lstm_7/mul_2:0"
+    name: "model_2/lstm_7/mul_2"
+    op_type: "Mul"
+    domain: ""
+  }
+  node {
+    input: "model_2/tf_op_layer_stack_4/stack_4:0"
+    output: "model_2/tf_op_layer_stack_5/stack_5_unsqueeze10"
+    name: "model_2/tf_op_layer_stack_5/stack_5_unsqueeze1"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: -1
+    }
+  }
+  node {
+    input: "model_2/lstm_7/mul_2:0"
+    output: "model_2/tf_op_layer_stack_3/stack_3_unsqueeze10"
+    name: "model_2/tf_op_layer_stack_3/stack_3_unsqueeze1"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 1
+    }
+  }
+  node {
+    input: "model_2/lstm_7/mul_2:0"
+    output: "model_2/lstm_7/stack_unsqueeze00"
+    name: "model_2/lstm_7/stack_unsqueeze0"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: 0
+    }
+  }
+  node {
+    input: "model_2/tf_op_layer_stack_3/stack_3_unsqueeze00"
+    input: "model_2/tf_op_layer_stack_3/stack_3_unsqueeze10"
+    output: "model_2/tf_op_layer_stack_3/stack_3:0"
+    name: "model_2/tf_op_layer_stack_3/stack_3_concat"
+    op_type: "Concat"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 1
+    }
+  }
+  node {
+    input: "model_2/lstm_7/stack_unsqueeze00"
+    output: "model_2/lstm_7/stack:0"
+    name: "model_2/lstm_7/stack_concat"
+    op_type: "Concat"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: 0
+    }
+  }
+  node {
+    input: "model_2/tf_op_layer_stack_3/stack_3:0"
+    output: "model_2/tf_op_layer_stack_5/stack_5_unsqueeze00"
+    name: "model_2/tf_op_layer_stack_5/stack_5_unsqueeze0"
+    op_type: "Unsqueeze"
+    domain: ""
+    attribute {
+      name: "axes"
+      type: INTS
+      ints: -1
+    }
+  }
+  node {
+    input: "model_2/lstm_7/stack:0"
+    output: "model_2/lstm_7/transpose_1:0"
+    name: "model_2/lstm_7/transpose_1"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 1
+      ints: 0
+      ints: 2
+    }
+  }
+  node {
+    input: "model_2/tf_op_layer_stack_5/stack_5_unsqueeze00"
+    input: "model_2/tf_op_layer_stack_5/stack_5_unsqueeze10"
+    output: "tf_op_layer_stack_5"
+    name: "model_2/tf_op_layer_stack_5/stack_5_concat"
+    op_type: "Concat"
+    domain: ""
+    attribute {
+      name: "axis"
+      type: INT
+      i: -1
+    }
+    doc_string: ""
+  }
+  node {
+    input: "model_2/lstm_7/transpose_1:0"
+    input: "shape_tensor1"
+    output: "model_2/dense_3/Tensordot/Reshape:0"
+    name: "model_2/dense_3/Tensordot/Reshape_reshape"
+    op_type: "Reshape"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/dense_3/Tensordot/Reshape:0"
+    input: "model_2/dense_3/Tensordot/Reshape_1:0"
+    output: "model_2/dense_3/Tensordot/MatMul:0"
+    name: "model_2/dense_3/Tensordot/MatMul_add"
+    op_type: "MatMul"
+    domain: ""
+  }
+  node {
+    input: "model_2/dense_3/Tensordot/MatMul:0"
+    input: "shape_tensor"
+    output: "model_2/dense_3/Tensordot:0"
+    name: "model_2/dense_3/Tensordot_reshape"
+    op_type: "Reshape"
+    domain: ""
+  }
+  node {
+    input: "model_2/dense_3/Tensordot:0"
+    input: "model_2/dense_3/BiasAdd/ReadVariableOp/resource:0"
+    output: "model_2/dense_3/BiasAdd:0"
+    name: "model_2/dense_3/BiasAdd_add"
+    op_type: "Add"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/dense_3/BiasAdd:0"
+    output: "model_2/activation_3/Sigmoid:0"
+    name: "model_2/activation_3/Sigmoid"
+    op_type: "Sigmoid"
+    domain: ""
+  }
+  node {
+    input: "transpose_output1"
+    input: "model_2/activation_3/Sigmoid:0"
+    output: "model_2/multiply_2/mul:0"
+    name: "model_2/multiply_2/mul"
+    op_type: "Mul"
+    domain: ""
+    doc_string: ""
+  }
+  node {
+    input: "model_2/multiply_2/mul:0"
+    input: "model_2/conv1d_3/Pad_pad_pads"
+    output: "model_2/conv1d_3/Pad_pad0"
+    name: "model_2/conv1d_3/Pad_pad"
+    op_type: "Pad"
+    domain: ""
+  }
+  node {
+    input: "model_2/conv1d_3/Pad_pad0"
+    output: "adjusted_input"
+    name: "Transpose"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 0
+      ints: 2
+      ints: 1
+    }
+    doc_string: ""
+  }
+  node {
+    input: "adjusted_input"
+    input: "conv1d_3/kernel:0"
+    output: "convolution_output"
+    name: "conv1d_3"
+    op_type: "Conv"
+    domain: ""
+    attribute {
+      name: "dilations"
+      type: INTS
+      ints: 1
+    }
+    attribute {
+      name: "group"
+      type: INT
+      i: 1
+    }
+    attribute {
+      name: "kernel_shape"
+      type: INTS
+      ints: 1
+    }
+    attribute {
+      name: "strides"
+      type: INTS
+      ints: 1
+    }
+    attribute {
+      name: "auto_pad"
+      type: STRING
+      s: "NOTSET"
+    }
+    attribute {
+      name: "pads"
+      type: INTS
+      ints: 0
+      ints: 0
+    }
+    doc_string: ""
+  }
+  node {
+    input: "convolution_output"
+    output: "conv1d_3"
+    name: "PushTranspose_1"
+    op_type: "Transpose"
+    domain: ""
+    attribute {
+      name: "perm"
+      type: INTS
+      ints: 0
+      ints: 2
+      ints: 1
+    }
+    doc_string: ""
+  }
+  initializer {
+      dims: 512
+      dims: 256
+      dims: 1
+      data_type: 1
+      name: "conv1d_3/kernel:0"
+  }
+  initializer {
+      dims: 6
+      data_type: 7
+      name: "model_2/conv1d_3/Pad_pad_pads"
+  }
+  initializer {
+      dims: 256
+      dims: 512
+      dims: 1
+      data_type: 1
+      name: "conv1d_2/kernel:0"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_start"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_end"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_axes"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_steps"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_start1"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_end1"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_axes1"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_steps1"
+  }
+  initializer {
+      dims: 3
+      data_type: 7
+      name: "shape_tensor"
+  }
+  initializer {
+      dims: 256
+      data_type: 1
+      name: "model_2/dense_3/BiasAdd/ReadVariableOp/resource:0"
+  }
+  initializer {
+      dims: 2
+      data_type: 7
+      name: "shape_tensor1"
+  }
+  initializer {
+      dims: 512
+      data_type: 1
+      name: "model_2/lstm_6/BiasAdd/ReadVariableOp/resource:0"
+  }
+  initializer {
+      dims: 512
+      data_type: 1
+      name: "model_2/lstm_7/BiasAdd/ReadVariableOp/resource:0"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_start2"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_end2"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_axes2"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_steps2"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_start3"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_end3"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_axes3"
+  }
+  initializer {
+      dims: 4
+      data_type: 7
+      name: "input_5:01_cropping_steps3"
+  }
+  initializer {
+      dims: 256
+      dims: 512
+      data_type: 1
+      name: "model_2/lstm_6/MatMul/ReadVariableOp/resource:0"
+  }
+  initializer {
+      dims: 128
+      dims: 512
+      data_type: 1
+      name: "model_2/lstm_6/MatMul_1/ReadVariableOp/resource:0"
+  }
+  initializer {
+      dims: 128
+      dims: 512
+      data_type: 1
+      name: "model_2/lstm_7/MatMul/ReadVariableOp/resource:0"
+  }
+  initializer {
+      dims: 128
+      dims: 512
+      data_type: 1
+      name: "model_2/lstm_7/MatMul_1/ReadVariableOp/resource:0"
+  }
+  initializer {
+      dims: 256
+      data_type: 1
+      name: "model_2/instant_layer_normalization_1/add_1/ReadVariableOp/resource:0"
+  }
+  initializer {
+      dims: 256
+      data_type: 1
+      name: "model_2/instant_layer_normalization_1/mul/ReadVariableOp/resource:0"
+  }
+  initializer {
+      data_type: 1
+      name: "model_2/instant_layer_normalization_1/add/y:0"
+  }
+  initializer {
+      dims: 128
+      dims: 256
+      data_type: 1
+      name: "model_2/dense_3/Tensordot/Reshape_1:0"
+  }
+  input {
+    name: "input_4"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 512
+          }
+        }
+      }
+    }
+  }
+  input {
+    name: "input_5"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 128
+          }
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "conv1d_3"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_param: "N"
+          }
+          dim {
+            dim_value: 1
+          }
+          dim {
+            dim_value: 512
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "tf_op_layer_stack_5"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_param: "N"
+          }
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 128
+          }
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+}
+opset_import {
+  version: 11
+}

models/ailia-models/source.txt

@@ -0,0 +1,7 @@
+https://github.com/axinc-ai/ailia-models/tree/master/audio_processing/dtln
+
+https://storage.googleapis.com/ailia-models/dtln/dtln1.onnx
+https://storage.googleapis.com/ailia-models/dtln/dtln1.onnx.prototxt
+
+https://storage.googleapis.com/ailia-models/dtln/dtln2.onnx
+https://storage.googleapis.com/ailia-models/dtln/dtln2.onnx.prototxt