From c20c871e9f963151fa410dd616c6b23d001ecdd2 Mon Sep 17 00:00:00 2001
From: Xian Shi <40013335+R1ckShi@users.noreply.github.com>
Date: 星期二, 04 七月 2023 19:57:04 +0800
Subject: [PATCH] Merge pull request #673 from alibaba-damo-academy/dev_clas
---
funasr/bin/asr_inference_launch.py | 22 ++++++++++++----------
1 files changed, 12 insertions(+), 10 deletions(-)
diff --git a/funasr/bin/asr_inference_launch.py b/funasr/bin/asr_inference_launch.py
index 656a965..a752f29 100644
--- a/funasr/bin/asr_inference_launch.py
+++ b/funasr/bin/asr_inference_launch.py
@@ -19,8 +19,8 @@
import numpy as np
import torch
import torchaudio
+import soundfile
import yaml
-from typeguard import check_argument_types
from funasr.bin.asr_infer import Speech2Text
from funasr.bin.asr_infer import Speech2TextMFCCA
@@ -79,7 +79,6 @@
param_dict: dict = None,
**kwargs,
):
- assert check_argument_types()
ncpu = kwargs.get("ncpu", 1)
torch.set_num_threads(ncpu)
if batch_size > 1:
@@ -239,7 +238,6 @@
param_dict: dict = None,
**kwargs,
):
- assert check_argument_types()
ncpu = kwargs.get("ncpu", 1)
torch.set_num_threads(ncpu)
@@ -259,6 +257,7 @@
export_mode = param_dict.get("export_mode", False)
else:
hotword_list_or_file = None
+ clas_scale = param_dict.get('clas_scale', 1.0)
if kwargs.get("device", None) == "cpu":
ngpu = 0
@@ -291,6 +290,7 @@
penalty=penalty,
nbest=nbest,
hotword_list_or_file=hotword_list_or_file,
+ clas_scale=clas_scale,
)
speech2text = Speech2TextParaformer(**speech2text_kwargs)
@@ -480,7 +480,6 @@
param_dict: dict = None,
**kwargs,
):
- assert check_argument_types()
ncpu = kwargs.get("ncpu", 1)
torch.set_num_threads(ncpu)
@@ -748,7 +747,6 @@
param_dict: dict = None,
**kwargs,
):
- assert check_argument_types()
if word_lm_train_config is not None:
raise NotImplementedError("Word LM is not implemented")
@@ -863,7 +861,13 @@
raw_inputs = _load_bytes(data_path_and_name_and_type[0])
raw_inputs = torch.tensor(raw_inputs)
if data_path_and_name_and_type is not None and data_path_and_name_and_type[2] == "sound":
- raw_inputs = torchaudio.load(data_path_and_name_and_type[0])[0][0]
+ try:
+ raw_inputs = torchaudio.load(data_path_and_name_and_type[0])[0][0]
+ except:
+ raw_inputs = soundfile.read(data_path_and_name_and_type[0], dtype='float32')[0]
+ if raw_inputs.ndim == 2:
+ raw_inputs = raw_inputs[:, 0]
+ raw_inputs = torch.tensor(raw_inputs)
if data_path_and_name_and_type is None and raw_inputs is not None:
if isinstance(raw_inputs, np.ndarray):
raw_inputs = torch.tensor(raw_inputs)
@@ -950,7 +954,6 @@
param_dict: dict = None,
**kwargs,
):
- assert check_argument_types()
ncpu = kwargs.get("ncpu", 1)
torch.set_num_threads(ncpu)
if batch_size > 1:
@@ -1119,7 +1122,6 @@
param_dict: dict = None,
**kwargs,
):
- assert check_argument_types()
ncpu = kwargs.get("ncpu", 1)
torch.set_num_threads(ncpu)
if batch_size > 1:
@@ -1307,7 +1309,6 @@
right_context: Number of frames in right context AFTER subsampling.
display_partial_hypotheses: Whether to display partial hypotheses.
"""
- assert check_argument_types()
if batch_size > 1:
raise NotImplementedError("batch decoding is not implemented")
@@ -1457,7 +1458,6 @@
param_dict: dict = None,
**kwargs,
):
- assert check_argument_types()
if batch_size > 1:
raise NotImplementedError("batch decoding is not implemented")
if word_lm_train_config is not None:
@@ -1606,6 +1606,8 @@
return inference_mfcca(**kwargs)
elif mode == "rnnt":
return inference_transducer(**kwargs)
+ elif mode == "bat":
+ return inference_transducer(**kwargs)
elif mode == "sa_asr":
return inference_sa_asr(**kwargs)
else:
--
Gitblit v1.9.1