From 28ccfbfc51068a663a80764e14074df5edf2b5ba Mon Sep 17 00:00:00 2001
From: kongdeqiang <kongdeqiang960204@163.com>
Date: 星期五, 13 三月 2026 17:41:41 +0800
Subject: [PATCH] 提交
---
runtime/python/onnxruntime/funasr_onnx/vad_bin.py | 51 ++++++++++++++++++++++++++++++++++-----------------
1 files changed, 34 insertions(+), 17 deletions(-)
diff --git a/runtime/python/onnxruntime/funasr_onnx/vad_bin.py b/runtime/python/onnxruntime/funasr_onnx/vad_bin.py
index 92928a8..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
@@ -69,7 +69,7 @@
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.vad_scorer_config = 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"]
)
@@ -90,10 +90,9 @@
waveform_list = self.load_data(audio_in, self.frontend.opts.frame_opts.samp_freq)
waveform_nums = len(waveform_list)
is_final = kwargs.get("kwargs", False)
-
segments = [[]] * self.batch_size
for beg_idx in range(0, waveform_nums, self.batch_size):
-
+ vad_scorer = E2EVadModel(self.vad_scorer_config)
end_idx = min(waveform_nums, beg_idx + self.batch_size)
waveform = waveform_list[beg_idx:end_idx]
feats, feats_len = self.extract_feat(waveform)
@@ -122,7 +121,7 @@
inputs.extend(in_cache)
scores, out_caches = self.infer(inputs)
in_cache = out_caches
- segments_part = self.vad_scorer(
+ segments_part = vad_scorer(
scores,
waveform_package,
is_final=is_final,
@@ -143,7 +142,24 @@
return segments
def load_data(self, wav_content: Union[str, np.ndarray, List[str]], fs: int = None) -> List:
+
+ def convert_to_wav(input_path, output_path):
+ from pydub import AudioSegment
+ try:
+ audio = AudioSegment.from_mp3(input_path)
+ audio.export(output_path, format="wav")
+ print("闊抽鏂囦欢涓簃p3鏍煎紡锛屽凡杞崲涓簑av鏍煎紡")
+
+ except Exception as e:
+ print(f"杞崲澶辫触:{e}")
+
def load_wav(path: str) -> np.ndarray:
+ if not path.lower().endswith('.wav'):
+ import os
+ input_path = path
+ path = os.path.splitext(path)[0]+'.wav'
+ convert_to_wav(input_path,path) #灏唌p3鏍煎紡杞崲鎴恮av鏍煎紡
+
waveform, _ = librosa.load(path, sr=fs)
return waveform
@@ -231,19 +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:
@@ -259,20 +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
)
@@ -280,6 +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:
@@ -299,13 +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