From f5d84f3194cc56dc363678d8018e563f0ce7e9e7 Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期五, 24 三月 2023 13:55:39 +0800
Subject: [PATCH] Merge branch 'main' of github.com:alibaba-damo-academy/FunASR add
---
funasr/export/export_model.py | 16 +++++++++-------
1 files changed, 9 insertions(+), 7 deletions(-)
diff --git a/funasr/export/export_model.py b/funasr/export/export_model.py
index 7e65a8f..b1161cb 100644
--- a/funasr/export/export_model.py
+++ b/funasr/export/export_model.py
@@ -10,11 +10,11 @@
from funasr.export.models import get_model
import numpy as np
import random
-
+from funasr.utils.types import str2bool
# torch_version = float(".".join(torch.__version__.split(".")[:2]))
# assert torch_version > 1.9
-class ASRModelExportParaformer:
+class ModelExport:
def __init__(
self,
cache_dir: Union[Path, str] = None,
@@ -74,8 +74,9 @@
# using dummy inputs for a example
if self.audio_in is not None:
feats, feats_len = self.load_feats(self.audio_in)
- for feat, len in zip(feats, feats_len):
- m(feat, len)
+ for i, (feat, len) in enumerate(zip(feats, feats_len)):
+ with torch.no_grad():
+ m(feat, len)
else:
dummy_input = model.get_dummy_inputs()
m(*dummy_input)
@@ -148,7 +149,7 @@
feats = []
feats_len = []
for line in wav_list:
- name, path = line.strip().split()
+ path = line.strip()
waveform, sampling_rate = torchaudio.load(path)
if sampling_rate != self.frontend.fs:
waveform = torchaudio.transforms.Resample(orig_freq=sampling_rate,
@@ -184,6 +185,7 @@
model, asr_train_args = ASRTask.build_model_from_file(
asr_train_config, asr_model_file, cmvn_file, 'cpu'
)
+ self.frontend = model.frontend
self._export(model, tag_name)
@@ -232,13 +234,13 @@
parser.add_argument('--model-name', type=str, required=True)
parser.add_argument('--export-dir', type=str, required=True)
parser.add_argument('--type', type=str, default='onnx', help='["onnx", "torch"]')
- parser.add_argument('--quantize', action='store_true', help='export quantized model')
+ 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')
args = parser.parse_args()
- export_model = ASRModelExportParaformer(
+ export_model = ModelExport(
cache_dir=args.export_dir,
onnx=args.type == 'onnx',
quant=args.quantize,
--
Gitblit v1.9.1