From c542eacb0aadcbc49c63db40429fca4e08f807a4 Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期五, 21 七月 2023 10:27:35 +0800
Subject: [PATCH] Merge branch 'main' of github.com:alibaba-damo-academy/FunASR add
---
funasr/export/models/language_models/transformer.py | 110 ++++
funasr/modules/vgg2l.py | 92 ++++
egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/infer.sh | 103 ++++
funasr/export/models/language_models/embed.py | 403 +++++++++++++++++
funasr/export/models/language_models/subsampling.py | 185 ++++++++
egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/README.md | 1
setup.py | 2
egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/utils | 1
funasr/export/export_conformer.py | 151 ++++++
funasr/export/models/language_models/__init__.py | 0
funasr/export/models/__init__.py | 4
funasr/export/models/decoder/xformer_decoder.py | 121 +++++
egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/finetune.py | 36 +
funasr/export/models/e2e_asr_conformer.py | 69 +++
egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/infer.py | 1
egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/demo.py | 11
funasr/export/models/language_models/seq_rnn.py | 84 +++
17 files changed, 1,373 insertions(+), 1 deletions(-)
diff --git a/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/README.md b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/README.md
new file mode 120000
index 0000000..bb55ab5
--- /dev/null
+++ b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/README.md
@@ -0,0 +1 @@
+../../TEMPLATE/README.md
\ No newline at end of file
diff --git a/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/demo.py b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/demo.py
new file mode 100644
index 0000000..ea486c8
--- /dev/null
+++ b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/demo.py
@@ -0,0 +1,11 @@
+from modelscope.pipelines import pipeline
+from modelscope.utils.constant import Tasks
+
+inference_pipeline = pipeline(
+ task=Tasks.auto_speech_recognition,
+ model='damo/damo/speech_paraformer_asr-en-16k-vocab4199-pytorch',
+ model_revision="v1.0.1",
+)
+audio_in='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_en.wav'
+rec_result = inference_pipeline(audio_in=audio_in)
+print(rec_result)
diff --git a/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/finetune.py b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/finetune.py
new file mode 100644
index 0000000..a945b31
--- /dev/null
+++ b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/finetune.py
@@ -0,0 +1,36 @@
+import os
+
+from modelscope.metainfo import Trainers
+from modelscope.trainers import build_trainer
+
+from funasr.datasets.ms_dataset import MsDataset
+from funasr.utils.modelscope_param import modelscope_args
+
+
+def modelscope_finetune(params):
+ if not os.path.exists(params.output_dir):
+ os.makedirs(params.output_dir, exist_ok=True)
+ # dataset split ["train", "validation"]
+ ds_dict = MsDataset.load(params.data_path)
+ kwargs = dict(
+ model=params.model,
+ data_dir=ds_dict,
+ dataset_type=params.dataset_type,
+ work_dir=params.output_dir,
+ batch_bins=params.batch_bins,
+ max_epoch=params.max_epoch,
+ lr=params.lr)
+ trainer = build_trainer(Trainers.speech_asr_trainer, default_args=kwargs)
+ trainer.train()
+
+
+if __name__ == '__main__':
+ params = modelscope_args(model="damo/speech_paraformer_asr-en-16k-vocab4199-pytorch", data_path="./data")
+ params.output_dir = "./checkpoint" # m妯″瀷淇濆瓨璺緞
+ params.data_path = "./example_data/" # 鏁版嵁璺緞
+ params.dataset_type = "small" # 灏忔暟鎹噺璁剧疆small锛岃嫢鏁版嵁閲忓ぇ浜�1000灏忔椂锛岃浣跨敤large
+ params.batch_bins = 2000 # batch size锛屽鏋渄ataset_type="small"锛宐atch_bins鍗曚綅涓篺bank鐗瑰緛甯ф暟锛屽鏋渄ataset_type="large"锛宐atch_bins鍗曚綅涓烘绉掞紝
+ params.max_epoch = 50 # 鏈�澶ц缁冭疆鏁�
+ params.lr = 0.00005 # 璁剧疆瀛︿範鐜�
+
+ modelscope_finetune(params)
diff --git a/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/infer.py b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/infer.py
new file mode 120000
index 0000000..128fc31
--- /dev/null
+++ b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/infer.py
@@ -0,0 +1 @@
+../../TEMPLATE/infer.py
\ No newline at end of file
diff --git a/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/infer.sh b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/infer.sh
new file mode 100644
index 0000000..b552ae8
--- /dev/null
+++ b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/infer.sh
@@ -0,0 +1,103 @@
+#!/usr/bin/env bash
+
+set -e
+set -u
+set -o pipefail
+
+stage=1
+stop_stage=2
+model="damo/speech_paraformer_asr-en-16k-vocab4199-pytorch"
+data_dir="./data/test"
+output_dir="./results"
+batch_size=64
+gpu_inference=true # whether to perform gpu decoding
+gpuid_list="0,1" # set gpus, e.g., gpuid_list="0,1"
+njob=64 # the number of jobs for CPU decoding, if gpu_inference=false, use CPU decoding, please set njob
+checkpoint_dir=
+checkpoint_name="valid.cer_ctc.ave.pb"
+
+. utils/parse_options.sh || exit 1;
+
+if ${gpu_inference} == "true"; then
+ nj=$(echo $gpuid_list | awk -F "," '{print NF}')
+else
+ nj=$njob
+ batch_size=1
+ gpuid_list=""
+ for JOB in $(seq ${nj}); do
+ gpuid_list=$gpuid_list"-1,"
+ done
+fi
+
+mkdir -p $output_dir/split
+split_scps=""
+for JOB in $(seq ${nj}); do
+ split_scps="$split_scps $output_dir/split/wav.$JOB.scp"
+done
+perl utils/split_scp.pl ${data_dir}/wav.scp ${split_scps}
+
+if [ -n "${checkpoint_dir}" ]; then
+ python utils/prepare_checkpoint.py ${model} ${checkpoint_dir} ${checkpoint_name}
+ model=${checkpoint_dir}/${model}
+fi
+
+if [ $stage -le 1 ] && [ $stop_stage -ge 1 ];then
+ echo "Decoding ..."
+ gpuid_list_array=(${gpuid_list//,/ })
+ for JOB in $(seq ${nj}); do
+ {
+ id=$((JOB-1))
+ gpuid=${gpuid_list_array[$id]}
+ mkdir -p ${output_dir}/output.$JOB
+ python infer.py \
+ --model ${model} \
+ --audio_in ${output_dir}/split/wav.$JOB.scp \
+ --output_dir ${output_dir}/output.$JOB \
+ --batch_size ${batch_size} \
+ --gpuid ${gpuid}
+ }&
+ done
+ wait
+
+ mkdir -p ${output_dir}/1best_recog
+ for f in token score text; do
+ if [ -f "${output_dir}/output.1/1best_recog/${f}" ]; then
+ for i in $(seq "${nj}"); do
+ cat "${output_dir}/output.${i}/1best_recog/${f}"
+ done | sort -k1 >"${output_dir}/1best_recog/${f}"
+ fi
+ done
+fi
+
+if [ $stage -le 2 ] && [ $stop_stage -ge 2 ];then
+ echo "Computing WER ..."
+ cp ${output_dir}/1best_recog/text ${output_dir}/1best_recog/text.proc
+ cp ${data_dir}/text ${output_dir}/1best_recog/text.ref
+ python utils/compute_wer.py ${output_dir}/1best_recog/text.ref ${output_dir}/1best_recog/text.proc ${output_dir}/1best_recog/text.cer
+ tail -n 3 ${output_dir}/1best_recog/text.cer
+fi
+
+if [ $stage -le 3 ] && [ $stop_stage -ge 3 ];then
+ echo "SpeechIO TIOBE textnorm"
+ echo "$0 --> Normalizing REF text ..."
+ ./utils/textnorm_zh.py \
+ --has_key --to_upper \
+ ${data_dir}/text \
+ ${output_dir}/1best_recog/ref.txt
+
+ echo "$0 --> Normalizing HYP text ..."
+ ./utils/textnorm_zh.py \
+ --has_key --to_upper \
+ ${output_dir}/1best_recog/text.proc \
+ ${output_dir}/1best_recog/rec.txt
+ grep -v $'\t$' ${output_dir}/1best_recog/rec.txt > ${output_dir}/1best_recog/rec_non_empty.txt
+
+ echo "$0 --> computing WER/CER and alignment ..."
+ ./utils/error_rate_zh \
+ --tokenizer char \
+ --ref ${output_dir}/1best_recog/ref.txt \
+ --hyp ${output_dir}/1best_recog/rec_non_empty.txt \
+ ${output_dir}/1best_recog/DETAILS.txt | tee ${output_dir}/1best_recog/RESULTS.txt
+ rm -rf ${output_dir}/1best_recog/rec.txt ${output_dir}/1best_recog/rec_non_empty.txt
+fi
+
diff --git a/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/utils b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/utils
new file mode 120000
index 0000000..2ac163f
--- /dev/null
+++ b/egs_modelscope/asr/paraformer/speech_paraformer_asr-en-16k-vocab4199-pytorch/utils
@@ -0,0 +1 @@
+../../../../egs/aishell/transformer/utils
\ No newline at end of file
diff --git a/funasr/export/export_conformer.py b/funasr/export/export_conformer.py
new file mode 100644
index 0000000..4980775
--- /dev/null
+++ b/funasr/export/export_conformer.py
@@ -0,0 +1,151 @@
+import json
+from typing import Union, Dict
+from pathlib import Path
+
+import os
+import logging
+import torch
+
+from funasr.export.models import get_model
+import numpy as np
+import random
+from funasr.utils.types import str2bool, str2triple_str
+# torch_version = float(".".join(torch.__version__.split(".")[:2]))
+# assert torch_version > 1.9
+
+class ModelExport:
+ def __init__(
+ self,
+ cache_dir: Union[Path, str] = None,
+ onnx: bool = True,
+ device: str = "cpu",
+ quant: bool = True,
+ fallback_num: int = 0,
+ audio_in: str = None,
+ calib_num: int = 200,
+ model_revision: str = None,
+ ):
+ self.set_all_random_seed(0)
+
+ self.cache_dir = cache_dir
+ self.export_config = dict(
+ feats_dim=560,
+ onnx=False,
+ )
+
+ self.onnx = onnx
+ self.device = device
+ self.quant = quant
+ self.fallback_num = fallback_num
+ self.frontend = None
+ self.audio_in = audio_in
+ self.calib_num = calib_num
+ self.model_revision = model_revision
+
+ def _export(
+ self,
+ model,
+ model_dir: str = None,
+ verbose: bool = False,
+ ):
+
+ export_dir = model_dir
+ os.makedirs(export_dir, exist_ok=True)
+
+ self.export_config["model_name"] = "model"
+ model = get_model(
+ model,
+ self.export_config,
+ )
+ model.eval()
+
+ if self.onnx:
+ self._export_onnx(model, verbose, export_dir)
+
+ print("output dir: {}".format(export_dir))
+
+ def _export_onnx(self, model, verbose, path):
+ model._export_onnx(verbose, path)
+
+ def set_all_random_seed(self, seed: int):
+ random.seed(seed)
+ np.random.seed(seed)
+ torch.random.manual_seed(seed)
+
+ def parse_audio_in(self, audio_in):
+
+ wav_list, name_list = [], []
+ if audio_in.endswith(".scp"):
+ f = open(audio_in, 'r')
+ lines = f.readlines()[:self.calib_num]
+ for line in lines:
+ name, path = line.strip().split()
+ name_list.append(name)
+ wav_list.append(path)
+ else:
+ wav_list = [audio_in,]
+ name_list = ["test",]
+ return wav_list, name_list
+
+ def load_feats(self, audio_in: str = None):
+ import torchaudio
+
+ wav_list, name_list = self.parse_audio_in(audio_in)
+ feats = []
+ feats_len = []
+ for line in wav_list:
+ path = line.strip()
+ waveform, sampling_rate = torchaudio.load(path)
+ if sampling_rate != self.frontend.fs:
+ waveform = torchaudio.transforms.Resample(orig_freq=sampling_rate,
+ new_freq=self.frontend.fs)(waveform)
+ fbank, fbank_len = self.frontend(waveform, [waveform.size(1)])
+ feats.append(fbank)
+ feats_len.append(fbank_len)
+ return feats, feats_len
+
+ def export(self,
+ mode: str = None,
+ ):
+
+ if mode.startswith('conformer'):
+ from funasr.tasks.asr import ASRTask
+ config = os.path.join(model_dir, 'config.yaml')
+ model_file = os.path.join(model_dir, 'model.pb')
+ cmvn_file = os.path.join(model_dir, 'am.mvn')
+ model, asr_train_args = ASRTask.build_model_from_file(
+ config, model_file, cmvn_file, 'cpu'
+ )
+ self.frontend = model.frontend
+ self.export_config["feats_dim"] = 560
+
+ self._export(model, self.cache_dir)
+
+if __name__ == '__main__':
+ import argparse
+ parser = argparse.ArgumentParser()
+ # parser.add_argument('--model-name', type=str, required=True)
+ parser.add_argument('--model-name', type=str, action="append", required=True, default=[])
+ parser.add_argument('--export-dir', type=str, required=True)
+ parser.add_argument('--type', type=str, default='onnx', help='["onnx", "torch"]')
+ parser.add_argument('--device', type=str, default='cpu', help='["cpu", "cuda"]')
+ parser.add_argument('--quantize', type=str2bool, default=False, help='export quantized model')
+ parser.add_argument('--fallback-num', type=int, default=0, help='amp fallback number')
+ parser.add_argument('--audio_in', type=str, default=None, help='["wav", "wav.scp"]')
+ parser.add_argument('--calib_num', type=int, default=200, help='calib max num')
+ parser.add_argument('--model_revision', type=str, default=None, help='model_revision')
+ args = parser.parse_args()
+
+ export_model = ModelExport(
+ cache_dir=args.export_dir,
+ onnx=args.type == 'onnx',
+ device=args.device,
+ quant=args.quantize,
+ fallback_num=args.fallback_num,
+ audio_in=args.audio_in,
+ calib_num=args.calib_num,
+ model_revision=args.model_revision,
+ )
+ for model_name in args.model_name:
+ print("export model: {}".format(model_name))
+ export_model.export(model_name)
diff --git a/funasr/export/models/__init__.py b/funasr/export/models/__init__.py
index 0e3a782..6177119 100644
--- a/funasr/export/models/__init__.py
+++ b/funasr/export/models/__init__.py
@@ -1,6 +1,8 @@
from funasr.models.e2e_asr_paraformer import Paraformer, BiCifParaformer
from funasr.export.models.e2e_asr_paraformer import Paraformer as Paraformer_export
from funasr.export.models.e2e_asr_paraformer import BiCifParaformer as BiCifParaformer_export
+from funasr.export.models.e2e_asr_conformer import Conformer as Conformer_export
+
from funasr.models.e2e_vad import E2EVadModel
from funasr.export.models.e2e_vad import E2EVadModel as E2EVadModel_export
from funasr.models.target_delay_transformer import TargetDelayTransformer
@@ -14,6 +16,8 @@
return BiCifParaformer_export(model, **export_config)
elif isinstance(model, Paraformer):
return Paraformer_export(model, **export_config)
+ elif isinstance(model, Conformer_export):
+ return Conformer_export(model, **export_config)
elif isinstance(model, E2EVadModel):
return E2EVadModel_export(model, **export_config)
elif isinstance(model, PunctuationModel):
diff --git a/funasr/export/models/decoder/xformer_decoder.py b/funasr/export/models/decoder/xformer_decoder.py
new file mode 100644
index 0000000..15199aa
--- /dev/null
+++ b/funasr/export/models/decoder/xformer_decoder.py
@@ -0,0 +1,121 @@
+import os
+
+import torch
+import torch.nn as nn
+
+from funasr.modules.attention import MultiHeadedAttention
+
+from funasr.export.models.modules.decoder_layer import DecoderLayer as OnnxDecoderLayer
+from funasr.export.models.language_models.embed import Embedding
+from funasr.export.models.modules.multihead_att import \
+ OnnxMultiHeadedAttention
+
+from funasr.export.utils.torch_function import MakePadMask, subsequent_mask
+
+class XformerDecoder(nn.Module):
+ def __init__(self,
+ model,
+ max_seq_len = 512,
+ model_name = 'decoder',
+ onnx: bool = True,):
+ super().__init__()
+ self.embed = Embedding(model.embed, max_seq_len)
+ self.model = model
+ if onnx:
+ self.make_pad_mask = MakePadMask(max_seq_len, flip=False)
+ else:
+ self.make_pad_mask = subsequent_mask(max_seq_len, flip=False)
+
+ if isinstance(self.model.decoders[0].self_attn, MultiHeadedAttention):
+ self.num_heads = self.model.decoders[0].self_attn.h
+ self.hidden_size = self.model.decoders[0].self_attn.linear_out.out_features
+
+ # replace multi-head attention module into customized module.
+ for i, d in enumerate(self.model.decoders):
+ # d is DecoderLayer
+ if isinstance(d.self_attn, MultiHeadedAttention):
+ d.self_attn = OnnxMultiHeadedAttention(d.self_attn)
+ if isinstance(d.src_attn, MultiHeadedAttention):
+ d.src_attn = OnnxMultiHeadedAttention(d.src_attn)
+ self.model.decoders[i] = OnnxDecoderLayer(d)
+
+ self.model_name = model_name
+
+ def prepare_mask(self, mask):
+ mask_3d_btd = mask[:, :, None]
+ if len(mask.shape) == 2:
+ mask_4d_bhlt = 1 - mask[:, None, None, :]
+ elif len(mask.shape) == 3:
+ mask_4d_bhlt = 1 - mask[:, None, :]
+
+ mask_4d_bhlt = mask_4d_bhlt * -10000.0
+ return mask_3d_btd, mask_4d_bhlt
+
+ def forward(self,
+ tgt,
+ memory,
+ cache):
+
+ mask = subsequent_mask(tgt.size(-1)).unsqueeze(0) # (B, T)
+
+ x = self.embed(tgt)
+ mask = self.prepare_mask(mask)
+ new_cache = []
+ for c, decoder in zip(cache, self.model.decoders):
+ x, mask = decoder(x, mask, memory, None, c)
+ new_cache.append(x)
+ x = x[:, 1:, :]
+
+ if self.model.normalize_before:
+ y = self.model.after_norm(x[:, -1])
+ else:
+ y = x[:, -1]
+
+ if self.model.output_layer is not None:
+ y = torch.log_softmax(self.model.output_layer(y), dim=-1)
+ return y, new_cache
+
+ def get_dummy_inputs(self, enc_size):
+ tgt = torch.LongTensor([0]).unsqueeze(0)
+ memory = torch.randn(1, 100, enc_size)
+ cache_num = len(self.model.decoders)
+ cache = [
+ torch.zeros((1, 1, self.model.decoders[0].size))
+ for _ in range(cache_num)
+ ]
+ return (tgt, memory, cache)
+
+ def is_optimizable(self):
+ return True
+
+ def get_input_names(self):
+ cache_num = len(self.model.decoders)
+ return ["tgt", "memory"] + [
+ "cache_%d" % i for i in range(cache_num)
+ ]
+
+ def get_output_names(self):
+ cache_num = len(self.model.decoders)
+ return ["y"] + ["out_cache_%d" % i for i in range(cache_num)]
+
+ def get_dynamic_axes(self):
+ ret = {
+ "tgt": {0: "tgt_batch", 1: "tgt_length"},
+ "memory": {0: "memory_batch", 1: "memory_length"},
+ }
+ cache_num = len(self.model.decoders)
+ ret.update(
+ {
+ "cache_%d" % d: {0: "cache_%d_batch" % d, 2: "cache_%d_length" % d}
+ for d in range(cache_num)
+ }
+ )
+ return ret
+
+ def get_model_config(self, path):
+ return {
+ "dec_type": "XformerDecoder",
+ "model_path": os.path.join(path, f"{self.model_name}.onnx"),
+ "n_layers": len(self.model.decoders),
+ "odim": self.model.decoders[0].size,
+ }
diff --git a/funasr/export/models/e2e_asr_conformer.py b/funasr/export/models/e2e_asr_conformer.py
new file mode 100644
index 0000000..45feda5
--- /dev/null
+++ b/funasr/export/models/e2e_asr_conformer.py
@@ -0,0 +1,69 @@
+import os
+import logging
+import torch
+import torch.nn as nn
+
+from funasr.export.utils.torch_function import MakePadMask
+from funasr.export.utils.torch_function import sequence_mask
+from funasr.models.encoder.conformer_encoder import ConformerEncoder
+from funasr.models.decoder.transformer_decoder import TransformerDecoder
+from funasr.export.models.encoder.conformer_encoder import ConformerEncoder as ConformerEncoder_export
+from funasr.export.models.decoder.xformer_decoder import XformerDecoder as TransformerDecoder_export
+
+class Conformer(nn.Module):
+ """
+ export conformer into onnx format
+ """
+
+ def __init__(
+ self,
+ model,
+ max_seq_len=512,
+ feats_dim=560,
+ model_name='model',
+ **kwargs,
+ ):
+ super().__init__()
+ onnx = False
+ if "onnx" in kwargs:
+ onnx = kwargs["onnx"]
+ if isinstance(model.encoder, ConformerEncoder):
+ self.encoder = ConformerEncoder_export(model.encoder, onnx=onnx)
+ elif isinstance(model.decoder, TransformerDecoder):
+ self.decoder = TransformerDecoder_export(model.decoder, onnx=onnx)
+
+ self.feats_dim = feats_dim
+ self.model_name = model_name
+
+ if onnx:
+ self.make_pad_mask = MakePadMask(max_seq_len, flip=False)
+ else:
+ self.make_pad_mask = sequence_mask(max_seq_len, flip=False)
+
+ def _export_model(self, model, verbose, path):
+ dummy_input = model.get_dummy_inputs()
+ model_script = model
+ model_path = os.path.join(path, f'{model.model_name}.onnx')
+ if not os.path.exists(model_path):
+ torch.onnx.export(
+ model_script,
+ dummy_input,
+ model_path,
+ verbose=verbose,
+ opset_version=14,
+ input_names=model.get_input_names(),
+ output_names=model.get_output_names(),
+ dynamic_axes=model.get_dynamic_axes()
+ )
+
+ def _export_encoder_onnx(self, verbose, path):
+ model_encoder = self.encoder
+ self._export_model(model_encoder, verbose, path)
+
+ def _export_decoder_onnx(self, verbose, path):
+ model_decoder = self.decoder
+ self._export_model(model_decoder, verbose, path)
+
+ def _export_onnx(self, verbose, path):
+ self._export_encoder_onnx(verbose, path)
+ self._export_decoder_onnx(verbose, path)
\ No newline at end of file
diff --git a/funasr/export/models/language_models/__init__.py b/funasr/export/models/language_models/__init__.py
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/funasr/export/models/language_models/__init__.py
diff --git a/funasr/export/models/language_models/embed.py b/funasr/export/models/language_models/embed.py
new file mode 100644
index 0000000..57748f2
--- /dev/null
+++ b/funasr/export/models/language_models/embed.py
@@ -0,0 +1,403 @@
+"""Positional Encoding Module."""
+
+import math
+
+import torch
+import torch.nn as nn
+from funasr.modules.embedding import (
+ LegacyRelPositionalEncoding, PositionalEncoding, RelPositionalEncoding,
+ ScaledPositionalEncoding, StreamPositionalEncoding)
+from funasr.modules.subsampling import (
+ Conv2dSubsampling, Conv2dSubsampling2, Conv2dSubsampling6,
+ Conv2dSubsampling8)
+from funasr.modules.subsampling_without_posenc import \
+ Conv2dSubsamplingWOPosEnc
+
+from funasr.export.models.language_models.subsampling import (
+ OnnxConv2dSubsampling, OnnxConv2dSubsampling2, OnnxConv2dSubsampling6,
+ OnnxConv2dSubsampling8)
+
+
+def get_pos_emb(pos_emb, max_seq_len=512, use_cache=True):
+ if isinstance(pos_emb, LegacyRelPositionalEncoding):
+ return OnnxLegacyRelPositionalEncoding(pos_emb, max_seq_len, use_cache)
+ elif isinstance(pos_emb, ScaledPositionalEncoding):
+ return OnnxScaledPositionalEncoding(pos_emb, max_seq_len, use_cache)
+ elif isinstance(pos_emb, RelPositionalEncoding):
+ return OnnxRelPositionalEncoding(pos_emb, max_seq_len, use_cache)
+ elif isinstance(pos_emb, PositionalEncoding):
+ return OnnxPositionalEncoding(pos_emb, max_seq_len, use_cache)
+ elif isinstance(pos_emb, StreamPositionalEncoding):
+ return OnnxStreamPositionalEncoding(pos_emb, max_seq_len, use_cache)
+ elif (isinstance(pos_emb, nn.Sequential) and len(pos_emb) == 0) or (
+ isinstance(pos_emb, Conv2dSubsamplingWOPosEnc)
+ ):
+ return pos_emb
+ else:
+ raise ValueError("Embedding model is not supported.")
+
+
+class Embedding(nn.Module):
+ def __init__(self, model, max_seq_len=512, use_cache=True):
+ super().__init__()
+ self.model = model
+ if not isinstance(model, nn.Embedding):
+ if isinstance(model, Conv2dSubsampling):
+ self.model = OnnxConv2dSubsampling(model)
+ self.model.out[-1] = get_pos_emb(model.out[-1], max_seq_len)
+ elif isinstance(model, Conv2dSubsampling2):
+ self.model = OnnxConv2dSubsampling2(model)
+ self.model.out[-1] = get_pos_emb(model.out[-1], max_seq_len)
+ elif isinstance(model, Conv2dSubsampling6):
+ self.model = OnnxConv2dSubsampling6(model)
+ self.model.out[-1] = get_pos_emb(model.out[-1], max_seq_len)
+ elif isinstance(model, Conv2dSubsampling8):
+ self.model = OnnxConv2dSubsampling8(model)
+ self.model.out[-1] = get_pos_emb(model.out[-1], max_seq_len)
+ else:
+ self.model[-1] = get_pos_emb(model[-1], max_seq_len)
+
+ def forward(self, x, mask=None):
+ if mask is None:
+ return self.model(x)
+ else:
+ return self.model(x, mask)
+
+
+def _pre_hook(
+ state_dict,
+ prefix,
+ local_metadata,
+ strict,
+ missing_keys,
+ unexpected_keys,
+ error_msgs,
+):
+ """Perform pre-hook in load_state_dict for backward compatibility.
+
+ Note:
+ We saved self.pe until v.0.5.2 but we have omitted it later.
+ Therefore, we remove the item "pe" from `state_dict` for backward compatibility.
+
+ """
+ k = prefix + "pe"
+ if k in state_dict:
+ state_dict.pop(k)
+
+
+class OnnxPositionalEncoding(torch.nn.Module):
+ """Positional encoding.
+
+ Args:
+ d_model (int): Embedding dimension.
+ dropout_rate (float): Dropout rate.
+ max_seq_len (int): Maximum input length.
+ reverse (bool): Whether to reverse the input position. Only for
+ the class LegacyRelPositionalEncoding. We remove it in the current
+ class RelPositionalEncoding.
+ """
+
+ def __init__(self, model, max_seq_len=512, reverse=False, use_cache=True):
+ """Construct an PositionalEncoding object."""
+ super(OnnxPositionalEncoding, self).__init__()
+ self.d_model = model.d_model
+ self.reverse = reverse
+ self.max_seq_len = max_seq_len
+ self.xscale = math.sqrt(self.d_model)
+ self._register_load_state_dict_pre_hook(_pre_hook)
+ self.pe = model.pe
+ self.use_cache = use_cache
+ self.model = model
+ if self.use_cache:
+ self.extend_pe()
+ else:
+ self.div_term = torch.exp(
+ torch.arange(0, self.d_model, 2, dtype=torch.float32)
+ * -(math.log(10000.0) / self.d_model)
+ )
+
+ def extend_pe(self):
+ """Reset the positional encodings."""
+ pe_length = len(self.pe[0])
+ if self.max_seq_len < pe_length:
+ self.pe = self.pe[:, : self.max_seq_len]
+ else:
+ self.model.extend_pe(torch.tensor(0.0).expand(1, self.max_seq_len))
+ self.pe = self.model.pe
+
+ def _add_pe(self, x):
+ """Computes positional encoding"""
+ if self.reverse:
+ position = torch.arange(
+ x.size(1) - 1, -1, -1.0, dtype=torch.float32
+ ).unsqueeze(1)
+ else:
+ position = torch.arange(0, x.size(1), dtype=torch.float32).unsqueeze(1)
+
+ x = x * self.xscale
+ x[:, :, 0::2] += torch.sin(position * self.div_term)
+ x[:, :, 1::2] += torch.cos(position * self.div_term)
+ return x
+
+ def forward(self, x: torch.Tensor):
+ """Add positional encoding.
+
+ Args:
+ x (torch.Tensor): Input tensor (batch, time, `*`).
+
+ Returns:
+ torch.Tensor: Encoded tensor (batch, time, `*`).
+ """
+ if self.use_cache:
+ x = x * self.xscale + self.pe[:, : x.size(1)]
+ else:
+ x = self._add_pe(x)
+ return x
+
+
+class OnnxScaledPositionalEncoding(OnnxPositionalEncoding):
+ """Scaled positional encoding module.
+
+ See Sec. 3.2 https://arxiv.org/abs/1809.08895
+
+ Args:
+ d_model (int): Embedding dimension.
+ dropout_rate (float): Dropout rate.
+ max_seq_len (int): Maximum input length.
+
+ """
+
+ def __init__(self, model, max_seq_len=512, use_cache=True):
+ """Initialize class."""
+ super().__init__(model, max_seq_len, use_cache=use_cache)
+ self.alpha = torch.nn.Parameter(torch.tensor(1.0))
+
+ def reset_parameters(self):
+ """Reset parameters."""
+ self.alpha.data = torch.tensor(1.0)
+
+ def _add_pe(self, x):
+ """Computes positional encoding"""
+ if self.reverse:
+ position = torch.arange(
+ x.size(1) - 1, -1, -1.0, dtype=torch.float32
+ ).unsqueeze(1)
+ else:
+ position = torch.arange(0, x.size(1), dtype=torch.float32).unsqueeze(1)
+
+ x = x * self.alpha
+ x[:, :, 0::2] += torch.sin(position * self.div_term)
+ x[:, :, 1::2] += torch.cos(position * self.div_term)
+ return x
+
+ def forward(self, x):
+ """Add positional encoding.
+
+ Args:
+ x (torch.Tensor): Input tensor (batch, time, `*`).
+
+ Returns:
+ torch.Tensor: Encoded tensor (batch, time, `*`).
+
+ """
+ if self.use_cache:
+ x = x + self.alpha * self.pe[:, : x.size(1)]
+ else:
+ x = self._add_pe(x)
+ return x
+
+
+class OnnxLegacyRelPositionalEncoding(OnnxPositionalEncoding):
+ """Relative positional encoding module (old version).
+
+ Details can be found in https://github.com/espnet/espnet/pull/2816.
+
+ See : Appendix B in https://arxiv.org/abs/1901.02860
+
+ Args:
+ d_model (int): Embedding dimension.
+ dropout_rate (float): Dropout rate.
+ max_seq_len (int): Maximum input length.
+
+ """
+
+ def __init__(self, model, max_seq_len=512, use_cache=True):
+ """Initialize class."""
+ super().__init__(model, max_seq_len, reverse=True, use_cache=use_cache)
+
+ def _get_pe(self, x):
+ """Computes positional encoding"""
+ if self.reverse:
+ position = torch.arange(
+ x.size(1) - 1, -1, -1.0, dtype=torch.float32
+ ).unsqueeze(1)
+ else:
+ position = torch.arange(0, x.size(1), dtype=torch.float32).unsqueeze(1)
+
+ pe = torch.zeros(x.shape)
+ pe[:, :, 0::2] += torch.sin(position * self.div_term)
+ pe[:, :, 1::2] += torch.cos(position * self.div_term)
+ return pe
+
+ def forward(self, x):
+ """Compute positional encoding.
+
+ Args:
+ x (torch.Tensor): Input tensor (batch, time, `*`).
+
+ Returns:
+ torch.Tensor: Encoded tensor (batch, time, `*`).
+ torch.Tensor: Positional embedding tensor (1, time, `*`).
+
+ """
+ x = x * self.xscale
+ if self.use_cache:
+ pos_emb = self.pe[:, : x.size(1)]
+ else:
+ pos_emb = self._get_pe(x)
+ return x, pos_emb
+
+
+class OnnxRelPositionalEncoding(torch.nn.Module):
+ """Relative positional encoding module (new implementation).
+ Details can be found in https://github.com/espnet/espnet/pull/2816.
+ See : Appendix B in https://arxiv.org/abs/1901.02860
+ Args:
+ d_model (int): Embedding dimension.
+ dropout_rate (float): Dropout rate.
+ max_seq_len (int): Maximum input length.
+ """
+
+ def __init__(self, model, max_seq_len=512, use_cache=True):
+ """Construct an PositionalEncoding object."""
+ super(OnnxRelPositionalEncoding, self).__init__()
+ self.d_model = model.d_model
+ self.xscale = math.sqrt(self.d_model)
+ self.pe = None
+ self.use_cache = use_cache
+ if self.use_cache:
+ self.extend_pe(torch.tensor(0.0).expand(1, max_seq_len))
+ else:
+ self.div_term = torch.exp(
+ torch.arange(0, self.d_model, 2, dtype=torch.float32)
+ * -(math.log(10000.0) / self.d_model)
+ )
+
+ def extend_pe(self, x):
+ """Reset the positional encodings."""
+ if self.pe is not None and self.pe.size(1) >= x.size(1) * 2 - 1:
+ # self.pe contains both positive and negative parts
+ # the length of self.pe is 2 * input_len - 1
+ if self.pe.dtype != x.dtype or self.pe.device != x.device:
+ self.pe = self.pe.to(dtype=x.dtype, device=x.device)
+ return
+ # Suppose `i` means to the position of query vecotr and `j` means the
+ # position of key vector. We use position relative positions when keys
+ # are to the left (i>j) and negative relative positions otherwise (i<j).
+ pe_positive = torch.zeros(x.size(1), self.d_model)
+ pe_negative = torch.zeros(x.size(1), self.d_model)
+ position = torch.arange(0, x.size(1), dtype=torch.float32).unsqueeze(1)
+ div_term = torch.exp(
+ torch.arange(0, self.d_model, 2, dtype=torch.float32)
+ * -(math.log(10000.0) / self.d_model)
+ )
+ pe_positive[:, 0::2] = torch.sin(position * div_term)
+ pe_positive[:, 1::2] = torch.cos(position * div_term)
+ pe_negative[:, 0::2] = torch.sin(-1 * position * div_term)
+ pe_negative[:, 1::2] = torch.cos(-1 * position * div_term)
+
+ # Reserve the order of positive indices and concat both positive and
+ # negative indices. This is used to support the shifting trick
+ # as in https://arxiv.org/abs/1901.02860
+ pe_positive = torch.flip(pe_positive, [0]).unsqueeze(0)
+ pe_negative = pe_negative[1:].unsqueeze(0)
+ pe = torch.cat([pe_positive, pe_negative], dim=1)
+ self.pe = pe.to(device=x.device, dtype=x.dtype)
+
+ def _get_pe(self, x):
+ pe_positive = torch.zeros(x.size(1), self.d_model)
+ pe_negative = torch.zeros(x.size(1), self.d_model)
+ theta = (
+ torch.arange(0, x.size(1), dtype=torch.float32).unsqueeze(1) * self.div_term
+ )
+ pe_positive[:, 0::2] = torch.sin(theta)
+ pe_positive[:, 1::2] = torch.cos(theta)
+ pe_negative[:, 0::2] = -1 * torch.sin(theta)
+ pe_negative[:, 1::2] = torch.cos(theta)
+
+ # Reserve the order of positive indices and concat both positive and
+ # negative indices. This is used to support the shifting trick
+ # as in https://arxiv.org/abs/1901.02860
+ pe_positive = torch.flip(pe_positive, [0]).unsqueeze(0)
+ pe_negative = pe_negative[1:].unsqueeze(0)
+ return torch.cat([pe_positive, pe_negative], dim=1)
+
+ def forward(self, x: torch.Tensor, use_cache=True):
+ """Add positional encoding.
+ Args:
+ x (torch.Tensor): Input tensor (batch, time, `*`).
+ Returns:
+ torch.Tensor: Encoded tensor (batch, time, `*`).
+ """
+ x = x * self.xscale
+ if self.use_cache:
+ pos_emb = self.pe[
+ :,
+ self.pe.size(1) // 2 - x.size(1) + 1 : self.pe.size(1) // 2 + x.size(1),
+ ]
+ else:
+ pos_emb = self._get_pe(x)
+ return x, pos_emb
+
+
+class OnnxStreamPositionalEncoding(torch.nn.Module):
+ """Streaming Positional encoding."""
+
+ def __init__(self, model, max_seq_len=5000, use_cache=True):
+ """Construct an PositionalEncoding object."""
+ super(StreamPositionalEncoding, self).__init__()
+ self.use_cache = use_cache
+ self.d_model = model.d_model
+ self.xscale = model.xscale
+ self.pe = model.pe
+ self.use_cache = use_cache
+ self.max_seq_len = max_seq_len
+ if self.use_cache:
+ self.extend_pe()
+ else:
+ self.div_term = torch.exp(
+ torch.arange(0, self.d_model, 2, dtype=torch.float32)
+ * -(math.log(10000.0) / self.d_model)
+ )
+ self._register_load_state_dict_pre_hook(_pre_hook)
+
+ def extend_pe(self):
+ """Reset the positional encodings."""
+ pe_length = len(self.pe[0])
+ if self.max_seq_len < pe_length:
+ self.pe = self.pe[:, : self.max_seq_len]
+ else:
+ self.model.extend_pe(self.max_seq_len)
+ self.pe = self.model.pe
+
+ def _add_pe(self, x, start_idx):
+ position = torch.arange(start_idx, x.size(1), dtype=torch.float32).unsqueeze(1)
+ x = x * self.xscale
+ x[:, :, 0::2] += torch.sin(position * self.div_term)
+ x[:, :, 1::2] += torch.cos(position * self.div_term)
+ return x
+
+ def forward(self, x: torch.Tensor, start_idx: int = 0):
+ """Add positional encoding.
+
+ Args:
+ x (torch.Tensor): Input tensor (batch, time, `*`).
+
+ Returns:
+ torch.Tensor: Encoded tensor (batch, time, `*`).
+
+ """
+ if self.use_cache:
+ return x * self.xscale + self.pe[:, start_idx : start_idx + x.size(1)]
+ else:
+ return self._add_pe(x, start_idx)
diff --git a/funasr/export/models/language_models/seq_rnn.py b/funasr/export/models/language_models/seq_rnn.py
new file mode 100644
index 0000000..ecff4b8
--- /dev/null
+++ b/funasr/export/models/language_models/seq_rnn.py
@@ -0,0 +1,84 @@
+import os
+
+import torch
+import torch.nn as nn
+
+class SequentialRNNLM(nn.Module):
+ def __init__(self, model, **kwargs):
+ super().__init__()
+ self.encoder = model.encoder
+ self.rnn = model.rnn
+ self.rnn_type = model.rnn_type
+ self.decoder = model.decoder
+ self.nlayers = model.nlayers
+ self.nhid = model.nhid
+ self.model_name = "seq_rnnlm"
+
+ def forward(self, y, hidden1, hidden2=None):
+ # batch_score function.
+ emb = self.encoder(y)
+ if self.rnn_type == "LSTM":
+ output, (hidden1, hidden2) = self.rnn(emb, (hidden1, hidden2))
+ else:
+ output, hidden1 = self.rnn(emb, hidden1)
+
+ decoded = self.decoder(
+ output.contiguous().view(output.size(0) * output.size(1), output.size(2))
+ )
+ if self.rnn_type == "LSTM":
+ return (
+ decoded.view(output.size(0), output.size(1), decoded.size(1)),
+ hidden1,
+ hidden2,
+ )
+ else:
+ return (
+ decoded.view(output.size(0), output.size(1), decoded.size(1)),
+ hidden1,
+ )
+
+ def get_dummy_inputs(self):
+ tgt = torch.LongTensor([0, 1]).unsqueeze(0)
+ hidden = torch.randn(self.nlayers, 1, self.nhid)
+ if self.rnn_type == "LSTM":
+ return (tgt, hidden, hidden)
+ else:
+ return (tgt, hidden)
+
+ def get_input_names(self):
+ if self.rnn_type == "LSTM":
+ return ["x", "in_hidden1", "in_hidden2"]
+ else:
+ return ["x", "in_hidden1"]
+
+ def get_output_names(self):
+ if self.rnn_type == "LSTM":
+ return ["y", "out_hidden1", "out_hidden2"]
+ else:
+ return ["y", "out_hidden1"]
+
+ def get_dynamic_axes(self):
+ ret = {
+ "x": {0: "x_batch", 1: "x_length"},
+ "y": {0: "y_batch"},
+ "in_hidden1": {1: "hidden1_batch"},
+ "out_hidden1": {1: "out_hidden1_batch"},
+ }
+ if self.rnn_type == "LSTM":
+ ret.update(
+ {
+ "in_hidden2": {1: "hidden2_batch"},
+ "out_hidden2": {1: "out_hidden2_batch"},
+ }
+ )
+ return ret
+
+ def get_model_config(self, path):
+ return {
+ "use_lm": True,
+ "model_path": os.path.join(path, f"{self.model_name}.onnx"),
+ "lm_type": "SequentialRNNLM",
+ "rnn_type": self.rnn_type,
+ "nhid": self.nhid,
+ "nlayers": self.nlayers,
+ }
diff --git a/funasr/export/models/language_models/subsampling.py b/funasr/export/models/language_models/subsampling.py
new file mode 100644
index 0000000..e71e127
--- /dev/null
+++ b/funasr/export/models/language_models/subsampling.py
@@ -0,0 +1,185 @@
+"""Subsampling layer definition."""
+
+import torch
+
+
+class OnnxConv2dSubsampling(torch.nn.Module):
+ """Convolutional 2D subsampling (to 1/4 length).
+
+ Args:
+ idim (int): Input dimension.
+ odim (int): Output dimension.
+ dropout_rate (float): Dropout rate.
+ pos_enc (torch.nn.Module): Custom position encoding layer.
+
+ """
+
+ def __init__(self, model):
+ """Construct an Conv2dSubsampling object."""
+ super().__init__()
+ self.conv = model.conv
+ self.out = model.out
+
+ def forward(self, x, x_mask):
+ """Subsample x.
+
+ Args:
+ x (torch.Tensor): Input tensor (#batch, time, idim).
+ x_mask (torch.Tensor): Input mask (#batch, 1, time).
+
+ Returns:
+ torch.Tensor: Subsampled tensor (#batch, time', odim),
+ where time' = time // 4.
+ torch.Tensor: Subsampled mask (#batch, 1, time'),
+ where time' = time // 4.
+
+ """
+ x = x.unsqueeze(1) # (b, c, t, f)
+ x = self.conv(x)
+ b, c, t, f = x.size()
+ x = self.out(x.transpose(1, 2).contiguous().view(b, t, c * f))
+ if x_mask is None:
+ return x, None
+ return x, x_mask[:, :-2:2][:, :-2:2]
+
+ def __getitem__(self, key):
+ """Get item.
+
+ When reset_parameters() is called, if use_scaled_pos_enc is used,
+ return the positioning encoding.
+
+ """
+ if key != -1:
+ raise NotImplementedError("Support only `-1` (for `reset_parameters`).")
+ return self.out[key]
+
+
+class OnnxConv2dSubsampling2(torch.nn.Module):
+ """Convolutional 2D subsampling (to 1/2 length).
+
+ Args:
+ idim (int): Input dimension.
+ odim (int): Output dimension.
+ dropout_rate (float): Dropout rate.
+ pos_enc (torch.nn.Module): Custom position encoding layer.
+
+ """
+
+ def __init__(self, model):
+ """Construct an Conv2dSubsampling object."""
+ super().__init__()
+ self.conv = model.conv
+ self.out = model.out
+
+ def forward(self, x, x_mask):
+ """Subsample x.
+
+ Args:
+ x (torch.Tensor): Input tensor (#batch, time, idim).
+ x_mask (torch.Tensor): Input mask (#batch, 1, time).
+
+ Returns:
+ torch.Tensor: Subsampled tensor (#batch, time', odim),
+ where time' = time // 2.
+ torch.Tensor: Subsampled mask (#batch, 1, time'),
+ where time' = time // 2.
+
+ """
+ x = x.unsqueeze(1) # (b, c, t, f)
+ x = self.conv(x)
+ b, c, t, f = x.size()
+ x = self.out(x.transpose(1, 2).contiguous().view(b, t, c * f))
+ if x_mask is None:
+ return x, None
+ return x, x_mask[:, :-2:2][:, :-2:1]
+
+ def __getitem__(self, key):
+ """Get item.
+
+ When reset_parameters() is called, if use_scaled_pos_enc is used,
+ return the positioning encoding.
+
+ """
+ if key != -1:
+ raise NotImplementedError("Support only `-1` (for `reset_parameters`).")
+ return self.out[key]
+
+
+class OnnxConv2dSubsampling6(torch.nn.Module):
+ """Convolutional 2D subsampling (to 1/6 length).
+
+ Args:
+ idim (int): Input dimension.
+ odim (int): Output dimension.
+ dropout_rate (float): Dropout rate.
+ pos_enc (torch.nn.Module): Custom position encoding layer.
+
+ """
+
+ def __init__(self, model):
+ """Construct an Conv2dSubsampling object."""
+ super().__init__()
+ self.conv = model.conv
+ self.out = model.out
+
+ def forward(self, x, x_mask):
+ """Subsample x.
+
+ Args:
+ x (torch.Tensor): Input tensor (#batch, time, idim).
+ x_mask (torch.Tensor): Input mask (#batch, 1, time).
+
+ Returns:
+ torch.Tensor: Subsampled tensor (#batch, time', odim),
+ where time' = time // 6.
+ torch.Tensor: Subsampled mask (#batch, 1, time'),
+ where time' = time // 6.
+
+ """
+ x = x.unsqueeze(1) # (b, c, t, f)
+ x = self.conv(x)
+ b, c, t, f = x.size()
+ x = self.out(x.transpose(1, 2).contiguous().view(b, t, c * f))
+ if x_mask is None:
+ return x, None
+ return x, x_mask[:, :-2:2][:, :-4:3]
+
+
+class OnnxConv2dSubsampling8(torch.nn.Module):
+ """Convolutional 2D subsampling (to 1/8 length).
+
+ Args:
+ idim (int): Input dimension.
+ odim (int): Output dimension.
+ dropout_rate (float): Dropout rate.
+ pos_enc (torch.nn.Module): Custom position encoding layer.
+
+ """
+
+ def __init__(self, model):
+ """Construct an Conv2dSubsampling object."""
+ super().__init__()
+ self.conv = model.conv
+ self.out = model.out
+
+ def forward(self, x, x_mask):
+ """Subsample x.
+
+ Args:
+ x (torch.Tensor): Input tensor (#batch, time, idim).
+ x_mask (torch.Tensor): Input mask (#batch, 1, time).
+
+ Returns:
+ torch.Tensor: Subsampled tensor (#batch, time', odim),
+ where time' = time // 8.
+ torch.Tensor: Subsampled mask (#batch, 1, time'),
+ where time' = time // 8.
+
+ """
+ x = x.unsqueeze(1) # (b, c, t, f)
+ x = self.conv(x)
+ b, c, t, f = x.size()
+ x = self.out(x.transpose(1, 2).contiguous().view(b, t, c * f))
+ if x_mask is None:
+ return x, None
+ return x, x_mask[:, :-2:2][:, :-2:2][:, :-2:2]
diff --git a/funasr/export/models/language_models/transformer.py b/funasr/export/models/language_models/transformer.py
new file mode 100644
index 0000000..ebf0574
--- /dev/null
+++ b/funasr/export/models/language_models/transformer.py
@@ -0,0 +1,110 @@
+import os
+
+import torch
+import torch.nn as nn
+from funasr.modules.vgg2l import import VGG2L
+from funasr.modules.attention import MultiHeadedAttention
+from funasr.modules.subsampling import (
+ Conv2dSubsampling, Conv2dSubsampling6, Conv2dSubsampling8)
+
+from funasr.export.models.modules.encoder_layer import EncoderLayerConformer as OnnxEncoderLayer
+from funasr.export.models.language_models.embed import Embedding
+from funasr.export.models.modules.multihead_att import OnnxMultiHeadedAttention
+
+from funasr.export.utils.torch_function import MakePadMask
+
+class TransformerLM(nn.Module, AbsExportModel):
+ def __init__(self, model, max_seq_len=512, **kwargs):
+ super().__init__()
+ self.embed = Embedding(model.embed, max_seq_len)
+ self.encoder = model.encoder
+ self.decoder = model.decoder
+ self.make_pad_mask = MakePadMask(max_seq_len, flip=False)
+ # replace multihead attention module into customized module.
+ for i, d in enumerate(self.encoder.encoders):
+ # d is EncoderLayer
+ if isinstance(d.self_attn, MultiHeadedAttention):
+ d.self_attn = OnnxMultiHeadedAttention(d.self_attn)
+ self.encoder.encoders[i] = OnnxEncoderLayer(d)
+
+ self.model_name = "transformer_lm"
+ self.num_heads = self.encoder.encoders[0].self_attn.h
+ self.hidden_size = self.encoder.encoders[0].self_attn.linear_out.out_features
+
+ def prepare_mask(self, mask):
+ if len(mask.shape) == 2:
+ mask = mask[:, None, None, :]
+ elif len(mask.shape) == 3:
+ mask = mask[:, None, :]
+ mask = 1 - mask
+ return mask * -10000.0
+
+ def forward(self, y, cache):
+ feats_length = torch.ones(y.shape).sum(dim=-1).type(torch.long)
+ mask = self.make_pad_mask(feats_length) # (B, T)
+ mask = (y != 0) * mask
+
+ xs = self.embed(y)
+ # forward_one_step of Encoder
+ if isinstance(
+ self.encoder.embed,
+ (Conv2dSubsampling, Conv2dSubsampling6, Conv2dSubsampling8, VGG2L),
+ ):
+ xs, mask = self.encoder.embed(xs, mask)
+ else:
+ xs = self.encoder.embed(xs)
+
+ new_cache = []
+ mask = self.prepare_mask(mask)
+ for c, e in zip(cache, self.encoder.encoders):
+ xs, mask = e(xs, mask, c)
+ new_cache.append(xs)
+
+ if self.encoder.normalize_before:
+ xs = self.encoder.after_norm(xs)
+
+ h = self.decoder(xs[:, -1])
+ return h, new_cache
+
+ def get_dummy_inputs(self):
+ tgt = torch.LongTensor([1]).unsqueeze(0)
+ cache = [
+ torch.zeros((1, 1, self.encoder.encoders[0].size))
+ for _ in range(len(self.encoder.encoders))
+ ]
+ return (tgt, cache)
+
+ def is_optimizable(self):
+ return True
+
+ def get_input_names(self):
+ return ["tgt"] + ["cache_%d" % i for i in range(len(self.encoder.encoders))]
+
+ def get_output_names(self):
+ return ["y"] + ["out_cache_%d" % i for i in range(len(self.encoder.encoders))]
+
+ def get_dynamic_axes(self):
+ ret = {"tgt": {0: "tgt_batch", 1: "tgt_length"}}
+ ret.update(
+ {
+ "cache_%d" % d: {0: "cache_%d_batch" % d, 1: "cache_%d_length" % d}
+ for d in range(len(self.encoder.encoders))
+ }
+ )
+ ret.update(
+ {
+ "out_cache_%d"
+ % d: {0: "out_cache_%d_batch" % d, 1: "out_cache_%d_length" % d}
+ for d in range(len(self.encoder.encoders))
+ }
+ )
+ return ret
+
+ def get_model_config(self, path):
+ return {
+ "use_lm": True,
+ "model_path": os.path.join(path, f"{self.model_name}.onnx"),
+ "lm_type": "TransformerLM",
+ "odim": self.encoder.encoders[0].size,
+ "nlayers": len(self.encoder.encoders),
+ }
diff --git a/funasr/modules/vgg2l.py b/funasr/modules/vgg2l.py
new file mode 100644
index 0000000..fb45d0b
--- /dev/null
+++ b/funasr/modules/vgg2l.py
@@ -0,0 +1,92 @@
+"""VGG2L module definition for custom encoder."""
+
+from typing import Tuple, Union
+
+import torch
+
+
+class VGG2L(torch.nn.Module):
+ """VGG2L module for custom encoder.
+
+ Args:
+ idim: Input dimension.
+ odim: Output dimension.
+ pos_enc: Positional encoding class.
+
+ """
+
+ def __init__(self, idim: int, odim: int, pos_enc: torch.nn.Module = None):
+ """Construct a VGG2L object."""
+ super().__init__()
+
+ self.vgg2l = torch.nn.Sequential(
+ torch.nn.Conv2d(1, 64, 3, stride=1, padding=1),
+ torch.nn.ReLU(),
+ torch.nn.Conv2d(64, 64, 3, stride=1, padding=1),
+ torch.nn.ReLU(),
+ torch.nn.MaxPool2d((3, 2)),
+ torch.nn.Conv2d(64, 128, 3, stride=1, padding=1),
+ torch.nn.ReLU(),
+ torch.nn.Conv2d(128, 128, 3, stride=1, padding=1),
+ torch.nn.ReLU(),
+ torch.nn.MaxPool2d((2, 2)),
+ )
+
+ if pos_enc is not None:
+ self.output = torch.nn.Sequential(
+ torch.nn.Linear(128 * ((idim // 2) // 2), odim), pos_enc
+ )
+ else:
+ self.output = torch.nn.Linear(128 * ((idim // 2) // 2), odim)
+
+ def forward(
+ self, feats: torch.Tensor, feats_mask: torch.Tensor
+ ) -> Union[
+ Tuple[torch.Tensor, torch.Tensor],
+ Tuple[Tuple[torch.Tensor, torch.Tensor], torch.Tensor],
+ ]:
+ """Forward VGG2L bottleneck.
+
+ Args:
+ feats: Feature sequences. (B, F, D_feats)
+ feats_mask: Mask of feature sequences. (B, 1, F)
+
+ Returns:
+ vgg_output: VGG output sequences.
+ (B, sub(F), D_out) or ((B, sub(F), D_out), (B, sub(F), D_att))
+ vgg_mask: Mask of VGG output sequences. (B, 1, sub(F))
+
+ """
+ feats = feats.unsqueeze(1)
+ vgg_output = self.vgg2l(feats)
+
+ b, c, t, f = vgg_output.size()
+
+ vgg_output = self.output(
+ vgg_output.transpose(1, 2).contiguous().view(b, t, c * f)
+ )
+
+ if feats_mask is not None:
+ vgg_mask = self.create_new_mask(feats_mask)
+ else:
+ vgg_mask = feats_mask
+
+ return vgg_output, vgg_mask
+
+ def create_new_mask(self, feats_mask: torch.Tensor) -> torch.Tensor:
+ """Create a subsampled mask of feature sequences.
+
+ Args:
+ feats_mask: Mask of feature sequences. (B, 1, F)
+
+ Returns:
+ vgg_mask: Mask of VGG2L output sequences. (B, 1, sub(F))
+
+ """
+ vgg1_t_len = feats_mask.size(2) - (feats_mask.size(2) % 3)
+ vgg_mask = feats_mask[:, :, :vgg1_t_len][:, :, ::3]
+
+ vgg2_t_len = vgg_mask.size(2) - (vgg_mask.size(2) % 2)
+ vgg_mask = vgg_mask[:, :, :vgg2_t_len][:, :, ::2]
+
+ return vgg_mask
diff --git a/setup.py b/setup.py
index 9c36dd0..43c734c 100644
--- a/setup.py
+++ b/setup.py
@@ -16,7 +16,7 @@
"librosa",
"jamo", # For kss
"PyYAML>=5.1.2",
- "soundfile>=0.10.2",
+ "soundfile>=0.12.1",
"h5py>=2.10.0",
"kaldiio>=2.17.0",
"torch_complex",
--
Gitblit v1.9.1