From c441eb08c44dfd4a7a8c68970fd3ebe7943d06ee Mon Sep 17 00:00:00 2001
From: shixian.shi <shixian.shi@alibaba-inc.com>
Date: 星期四, 09 三月 2023 15:26:03 +0800
Subject: [PATCH] tp_inference device bug
---
funasr/bin/tp_inference.py | 67 ++++++++++++++++++---------------
1 files changed, 36 insertions(+), 31 deletions(-)
diff --git a/funasr/bin/tp_inference.py b/funasr/bin/tp_inference.py
index 0fcffbb..67e82a7 100644
--- a/funasr/bin/tp_inference.py
+++ b/funasr/bin/tp_inference.py
@@ -87,18 +87,22 @@
else:
timestamp_list[-1][1] = num_frames*TIME_RATE
assert len(new_char_list) == len(timestamp_list)
- res = ""
+ res_str = ""
for char, timestamp in zip(new_char_list, timestamp_list):
- res += "{} {} {};".format(char, timestamp[0], timestamp[1])
- return res
+ res_str += "{} {} {};".format(char, str(timestamp[0]+0.0005)[:5], str(timestamp[1]+0.0005)[:5])
+ res = []
+ for char, timestamp in zip(char_list, timestamp_list):
+ if char != '<sil>':
+ res.append([int(timestamp[0] * 1000), int(timestamp[1] * 1000)])
+ return res_str, res
class SpeechText2Timestamp:
def __init__(
self,
- tp_train_config: Union[Path, str] = None,
- tp_model_file: Union[Path, str] = None,
- tp_cmvn_file: Union[Path, str] = None,
+ timestamp_infer_config: Union[Path, str] = None,
+ timestamp_model_file: Union[Path, str] = None,
+ timestamp_cmvn_file: Union[Path, str] = None,
device: str = "cpu",
dtype: str = "float32",
**kwargs,
@@ -106,11 +110,14 @@
assert check_argument_types()
# 1. Build ASR model
tp_model, tp_train_args = ASRTask.build_model_from_file(
- tp_train_config, tp_model_file, device
+ timestamp_infer_config, timestamp_model_file, device
)
+ if 'cuda' in device:
+ tp_model = tp_model.cuda()
+
frontend = None
if tp_train_args.frontend is not None:
- frontend = WavFrontend(cmvn_file=tp_cmvn_file, **tp_train_args.frontend_conf)
+ frontend = WavFrontend(cmvn_file=timestamp_cmvn_file, **tp_train_args.frontend_conf)
logging.info("tp_model: {}".format(tp_model))
logging.info("tp_train_args: {}".format(tp_train_args))
@@ -144,11 +151,11 @@
# Input as audio signal
if isinstance(speech, np.ndarray):
speech = torch.tensor(speech)
-
if self.frontend is not None:
feats, feats_len = self.frontend.forward(speech, speech_lengths)
feats = to_device(feats, device=self.device)
feats_len = feats_len.int()
+ self.tp_model.frontend = None
else:
feats = speech
feats_len = speech_lengths
@@ -174,9 +181,9 @@
ngpu: int,
log_level: Union[int, str],
data_path_and_name_and_type,
- tp_train_config: Optional[str],
- tp_model_file: Optional[str],
- tp_cmvn_file: Optional[str] = None,
+ timestamp_infer_config: Optional[str],
+ timestamp_model_file: Optional[str],
+ timestamp_cmvn_file: Optional[str] = None,
raw_inputs: Union[np.ndarray, torch.Tensor] = None,
key_file: Optional[str] = None,
allow_variable_data_keys: bool = False,
@@ -190,9 +197,9 @@
batch_size=batch_size,
ngpu=ngpu,
log_level=log_level,
- tp_train_config=tp_train_config,
- tp_model_file=tp_model_file,
- tp_cmvn_file=tp_cmvn_file,
+ timestamp_infer_config=timestamp_infer_config,
+ timestamp_model_file=timestamp_model_file,
+ timestamp_cmvn_file=timestamp_cmvn_file,
key_file=key_file,
allow_variable_data_keys=allow_variable_data_keys,
output_dir=output_dir,
@@ -209,9 +216,9 @@
ngpu: int,
log_level: Union[int, str],
# data_path_and_name_and_type,
- tp_train_config: Optional[str],
- tp_model_file: Optional[str],
- tp_cmvn_file: Optional[str] = None,
+ timestamp_infer_config: Optional[str],
+ timestamp_model_file: Optional[str],
+ timestamp_cmvn_file: Optional[str] = None,
# raw_inputs: Union[np.ndarray, torch.Tensor] = None,
key_file: Optional[str] = None,
allow_variable_data_keys: bool = False,
@@ -236,15 +243,14 @@
device = "cuda"
else:
device = "cpu"
-
# 1. Set random-seed
set_all_random_seed(seed)
# 2. Build speech2vadsegment
speechtext2timestamp_kwargs = dict(
- tp_train_config=tp_train_config,
- tp_model_file=tp_model_file,
- tp_cmvn_file=tp_cmvn_file,
+ timestamp_infer_config=timestamp_infer_config,
+ timestamp_model_file=timestamp_model_file,
+ timestamp_cmvn_file=timestamp_cmvn_file,
device=device,
dtype=dtype,
)
@@ -256,7 +262,8 @@
raw_inputs: Union[np.ndarray, torch.Tensor] = None,
output_dir_v2: Optional[str] = None,
fs: dict = None,
- param_dict: dict = None
+ param_dict: dict = None,
+ **kwargs
):
# 3. Build data-iterator
if data_path_and_name_and_type is None and raw_inputs is not None:
@@ -295,11 +302,9 @@
for batch_id in range(_bs):
key = keys[batch_id]
token = speechtext2timestamp.converter.ids2tokens(batch['text'][batch_id])
- timestamp = time_stamp_lfr6_advance(us_alphas[batch_id], us_cif_peak[batch_id], token)
- logging.warning(timestamp)
- import pdb; pdb.set_trace()
- tp_result_list.append({'text':"".join([i for i in token if i != '<sil>']), 'timestamp': timestamp})
-
+ ts_str, ts_list = time_stamp_lfr6_advance(us_alphas[batch_id], us_cif_peak[batch_id], token)
+ logging.warning(ts_str)
+ tp_result_list.append({'text':"".join([i for i in token if i != '<sil>']), 'timestamp': ts_list})
return tp_result_list
return _forward
@@ -362,17 +367,17 @@
group = parser.add_argument_group("The model configuration related")
group.add_argument(
- "--tp_train_config",
+ "--timestamp_infer_config",
type=str,
help="VAD infer configuration",
)
group.add_argument(
- "--tp_model_file",
+ "--timestamp_model_file",
type=str,
help="VAD model parameter file",
)
group.add_argument(
- "--tp_cmvn_file",
+ "--timestamp_cmvn_file",
type=str,
help="Global cmvn file",
)
--
Gitblit v1.9.1