From 9038340be707aa7cbf62fcbf33ab615bb266abdb Mon Sep 17 00:00:00 2001
From: 王梦迪 <73778524+di-osc@users.noreply.github.com>
Date: 星期一, 26 五月 2025 14:11:02 +0800
Subject: [PATCH] 修复Fsmn_vad_online多线程调用报错 (#2528)
---
runtime/python/onnxruntime/funasr_onnx/vad_bin.py | 27 +++++++------
runtime/python/onnxruntime/funasr_onnx/utils/frontend.py | 67 ++++++++++++++++++++-------------
2 files changed, 55 insertions(+), 39 deletions(-)
diff --git a/runtime/python/onnxruntime/funasr_onnx/utils/frontend.py b/runtime/python/onnxruntime/funasr_onnx/utils/frontend.py
index f4cc8ff..54f9deb 100644
--- a/runtime/python/onnxruntime/funasr_onnx/utils/frontend.py
+++ b/runtime/python/onnxruntime/funasr_onnx/utils/frontend.py
@@ -2,6 +2,7 @@
from pathlib import Path
from typing import Any, Dict, Iterable, List, NamedTuple, Set, Tuple, Union
import copy
+from functools import lru_cache
import numpy as np
import kaldi_native_fbank as knf
@@ -45,7 +46,7 @@
self.cmvn_file = cmvn_file
if self.cmvn_file:
- self.cmvn = self.load_cmvn()
+ self.cmvn = load_cmvn(self.cmvn_file)
self.fbank_fn = None
self.fbank_beg_idx = 0
self.reset_status()
@@ -122,33 +123,47 @@
inputs = (inputs + means) * vars
return inputs
- def load_cmvn(
- self,
- ) -> np.ndarray:
- with open(self.cmvn_file, "r", encoding="utf-8") as f:
- lines = f.readlines()
+@lru_cache()
+def load_cmvn(cmvn_file: Union[str, Path]) -> np.ndarray:
+ """load cmvn file to numpy array.
- means_list = []
- vars_list = []
- for i in range(len(lines)):
- line_item = lines[i].split()
- if line_item[0] == "<AddShift>":
- line_item = lines[i + 1].split()
- if line_item[0] == "<LearnRateCoef>":
- add_shift_line = line_item[3 : (len(line_item) - 1)]
- means_list = list(add_shift_line)
- continue
- elif line_item[0] == "<Rescale>":
- line_item = lines[i + 1].split()
- if line_item[0] == "<LearnRateCoef>":
- rescale_line = line_item[3 : (len(line_item) - 1)]
- vars_list = list(rescale_line)
- continue
+ Args:
+ cmvn_file (Union[str, Path]): cmvn file path.
- means = np.array(means_list).astype(np.float64)
- vars = np.array(vars_list).astype(np.float64)
- cmvn = np.array([means, vars])
- return cmvn
+ Raises:
+ FileNotFoundError: cmvn file not exits.
+
+ Returns:
+ np.ndarray: cmvn array. shape is (2, dim).The first row is means, the second row is vars.
+ """
+
+ cmvn_file = Path(cmvn_file)
+ if not cmvn_file.exists():
+ raise FileNotFoundError("cmvn file not exits")
+
+ with open(cmvn_file, "r", encoding="utf-8") as f:
+ lines = f.readlines()
+ means_list = []
+ vars_list = []
+ for i in range(len(lines)):
+ line_item = lines[i].split()
+ if line_item[0] == "<AddShift>":
+ line_item = lines[i + 1].split()
+ if line_item[0] == "<LearnRateCoef>":
+ add_shift_line = line_item[3 : (len(line_item) - 1)]
+ means_list = list(add_shift_line)
+ continue
+ elif line_item[0] == "<Rescale>":
+ line_item = lines[i + 1].split()
+ if line_item[0] == "<LearnRateCoef>":
+ rescale_line = line_item[3 : (len(line_item) - 1)]
+ vars_list = list(rescale_line)
+ continue
+
+ means = np.array(means_list).astype(np.float64)
+ vars = np.array(vars_list).astype(np.float64)
+ cmvn = np.array([means, vars])
+ return cmvn
class WavFrontendOnline(WavFrontend):
diff --git a/runtime/python/onnxruntime/funasr_onnx/vad_bin.py b/runtime/python/onnxruntime/funasr_onnx/vad_bin.py
index af4663a..f784f26 100644
--- a/runtime/python/onnxruntime/funasr_onnx/vad_bin.py
+++ b/runtime/python/onnxruntime/funasr_onnx/vad_bin.py
@@ -4,7 +4,7 @@
import os.path
from pathlib import Path
-from typing import List, Union, Tuple
+from typing import List, Union, Tuple, Dict
import copy
import librosa
@@ -247,19 +247,17 @@
model_dir = model.export(type="onnx", quantize=quantize, **kwargs)
config_file = os.path.join(model_dir, "config.yaml")
- cmvn_file = os.path.join(model_dir, "am.mvn")
- config = read_yaml(config_file)
+ self.cmvn_file = os.path.join(model_dir, "am.mvn")
+ self.config = read_yaml(config_file)
- self.frontend = WavFrontendOnline(cmvn_file=cmvn_file, **config["frontend_conf"])
self.ort_infer = OrtInferSession(
model_file, device_id, intra_op_num_threads=intra_op_num_threads
)
self.batch_size = batch_size
- self.vad_scorer = E2EVadModel(config["model_conf"])
self.max_end_sil = (
- max_end_sil if max_end_sil is not None else config["model_conf"]["max_end_silence_time"]
+ max_end_sil if max_end_sil is not None else self.config["model_conf"]["max_end_silence_time"]
)
- self.encoder_conf = config["encoder_conf"]
+ self.encoder_conf = self.config["encoder_conf"]
def prepare_cache(self, in_cache: list = []):
if len(in_cache) > 0:
@@ -275,20 +273,22 @@
def __call__(self, audio_in: np.ndarray, **kwargs) -> List:
waveforms = np.expand_dims(audio_in, axis=0)
- param_dict = kwargs.get("param_dict", dict())
+ param_dict: Dict = kwargs.get("param_dict", dict())
is_final = param_dict.get("is_final", False)
- feats, feats_len = self.extract_feat(waveforms, is_final)
+ frontend: WavFrontendOnline = param_dict.get("frontend", WavFrontendOnline(cmvn_file=self.cmvn_file, **self.config["frontend_conf"]))
+ feats, feats_len = self.extract_feat(frontend=frontend, waveforms=waveforms, is_final=is_final)
segments = []
if feats.size != 0:
in_cache = param_dict.get("in_cache", list())
in_cache = self.prepare_cache(in_cache)
+ vad_scorer = param_dict.get("vad_scorer", E2EVadModel(self.config["model_conf"]))
try:
inputs = [feats]
inputs.extend(in_cache)
scores, out_caches = self.infer(inputs)
param_dict["in_cache"] = out_caches
- waveforms = self.frontend.get_waveforms()
- segments = self.vad_scorer(
+ waveforms = frontend.get_waveforms()
+ segments = vad_scorer(
scores, waveforms, is_final=is_final, max_end_sil=self.max_end_sil, online=True
)
@@ -296,6 +296,7 @@
# logging.warning(traceback.format_exc())
logging.warning("input wav is silence or noise")
segments = []
+ param_dict.update({"frontend": frontend, "vad_scorer": vad_scorer})
return segments
def load_data(self, wav_content: Union[str, np.ndarray, List[str]], fs: int = None) -> List:
@@ -315,13 +316,13 @@
raise TypeError(f"The type of {wav_content} is not in [str, np.ndarray, list]")
def extract_feat(
- self, waveforms: np.ndarray, is_final: bool = False
+ self, frontend: WavFrontendOnline, waveforms: np.ndarray, is_final: bool = False
) -> Tuple[np.ndarray, np.ndarray]:
waveforms_lens = np.zeros(waveforms.shape[0]).astype(np.int32)
for idx, waveform in enumerate(waveforms):
waveforms_lens[idx] = waveform.shape[-1]
- feats, feats_len = self.frontend.extract_fbank(waveforms, waveforms_lens, is_final)
+ feats, feats_len = frontend.extract_fbank(waveforms, waveforms_lens, is_final)
# feats.append(feat)
# feats_len.append(feat_len)
--
Gitblit v1.9.1