From 4ace5a95b052d338947fc88809a440ccd55cf6b4 Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期四, 16 十一月 2023 16:39:52 +0800
Subject: [PATCH] funasr pages
---
funasr/utils/timestamp_tools.py | 155 ++++++++++++++++++++++++++++++++++++++-------------
1 files changed, 116 insertions(+), 39 deletions(-)
diff --git a/funasr/utils/timestamp_tools.py b/funasr/utils/timestamp_tools.py
index 73f0c7a..6594273 100644
--- a/funasr/utils/timestamp_tools.py
+++ b/funasr/utils/timestamp_tools.py
@@ -1,39 +1,61 @@
-from scipy.fftpack import shift
import torch
-import copy
import codecs
import logging
-import edit_distance
import argparse
import numpy as np
-from typing import Any, List, Tuple, Union
+import edit_distance
+from itertools import zip_longest
+
+
+def cif_wo_hidden(alphas, threshold):
+ batch_size, len_time = alphas.size()
+ # loop varss
+ integrate = torch.zeros([batch_size], device=alphas.device)
+ # intermediate vars along time
+ list_fires = []
+ for t in range(len_time):
+ alpha = alphas[:, t]
+ integrate += alpha
+ list_fires.append(integrate)
+ fire_place = integrate >= threshold
+ integrate = torch.where(fire_place,
+ integrate - torch.ones([batch_size], device=alphas.device)*threshold,
+ integrate)
+ fires = torch.stack(list_fires, 1)
+ return fires
def ts_prediction_lfr6_standard(us_alphas,
us_peaks,
char_list,
vad_offset=0.0,
- force_time_shift=-1.5
+ force_time_shift=-1.5,
+ sil_in_str=True
):
if not len(char_list):
- return []
+ return "", []
START_END_THRESHOLD = 5
MAX_TOKEN_DURATION = 12
TIME_RATE = 10.0 * 6 / 1000 / 3 # 3 times upsampled
if len(us_alphas.shape) == 2:
- _, peaks = us_alphas[0], us_peaks[0] # support inference batch_size=1 only
+ alphas, peaks = us_alphas[0], us_peaks[0] # support inference batch_size=1 only
else:
- _, peaks = us_alphas, us_peaks
- num_frames = peaks.shape[0]
+ alphas, peaks = us_alphas, us_peaks
if char_list[-1] == '</s>':
char_list = char_list[:-1]
+ fire_place = torch.where(peaks>1.0-1e-4)[0].cpu().numpy() + force_time_shift # total offset
+ if len(fire_place) != len(char_list) + 1:
+ alphas /= (alphas.sum() / (len(char_list) + 1))
+ alphas = alphas.unsqueeze(0)
+ peaks = cif_wo_hidden(alphas, threshold=1.0-1e-4)[0]
+ fire_place = torch.where(peaks>1.0-1e-4)[0].cpu().numpy() + force_time_shift # total offset
+ num_frames = peaks.shape[0]
timestamp_list = []
new_char_list = []
# for bicif model trained with large data, cif2 actually fires when a character starts
# so treat the frames between two peaks as the duration of the former token
fire_place = torch.where(peaks>1.0-1e-4)[0].cpu().numpy() + force_time_shift # total offset
- num_peak = len(fire_place)
- assert num_peak == len(char_list) + 1 # number of peaks is supposed to be number of tokens + 1
+ # assert num_peak == len(char_list) + 1 # number of peaks is supposed to be number of tokens + 1
# begin silence
if fire_place[0] > START_END_THRESHOLD:
# char_list.insert(0, '<sil>')
@@ -66,6 +88,8 @@
timestamp_list[i][1] = timestamp_list[i][1] + vad_offset / 1000.0
res_txt = ""
for char, timestamp in zip(new_char_list, timestamp_list):
+ #if char != '<sil>':
+ if not sil_in_str and char == '<sil>': continue
res_txt += "{} {} {};".format(char, str(timestamp[0]+0.0005)[:5], str(timestamp[1]+0.0005)[:5])
res = []
for char, timestamp in zip(new_char_list, timestamp_list):
@@ -75,6 +99,7 @@
def time_stamp_sentence(punc_id_list, time_stamp_postprocessed, text_postprocessed):
+ punc_list = ['锛�', '銆�', '锛�', '銆�']
res = []
if text_postprocessed is None:
return res
@@ -84,44 +109,54 @@
return res
if len(text_postprocessed) == 0:
return res
+
if punc_id_list is None or len(punc_id_list) == 0:
res.append({
'text': text_postprocessed.split(),
"start": time_stamp_postprocessed[0][0],
- "end": time_stamp_postprocessed[-1][1]
+ "end": time_stamp_postprocessed[-1][1],
+ 'text_seg': text_postprocessed.split(),
+ "ts_list": time_stamp_postprocessed,
})
return res
if len(punc_id_list) != len(time_stamp_postprocessed):
- res.append({
- 'text': text_postprocessed.split(),
- "start": time_stamp_postprocessed[0][0],
- "end": time_stamp_postprocessed[-1][1]
- })
- return res
-
- sentence_text = ''
+ print(" warning length mistach!!!!!!")
+ sentence_text = ""
+ sentence_text_seg = ""
+ ts_list = []
sentence_start = time_stamp_postprocessed[0][0]
+ sentence_end = time_stamp_postprocessed[0][1]
texts = text_postprocessed.split()
- for i in range(len(punc_id_list)):
- sentence_text += texts[i]
- if punc_id_list[i] == 2:
- sentence_text += ','
+ punc_stamp_text_list = list(zip_longest(punc_id_list, time_stamp_postprocessed, texts, fillvalue=None))
+ for punc_stamp_text in punc_stamp_text_list:
+ punc_id, time_stamp, text = punc_stamp_text
+ # sentence_text += text if text is not None else ''
+ if text is not None:
+ if 'a' <= text[0] <= 'z' or 'A' <= text[0] <= 'Z':
+ sentence_text += ' ' + text
+ elif len(sentence_text) and ('a' <= sentence_text[-1] <= 'z' or 'A' <= sentence_text[-1] <= 'Z'):
+ sentence_text += ' ' + text
+ else:
+ sentence_text += text
+ sentence_text_seg += text + ' '
+ ts_list.append(time_stamp)
+
+ punc_id = int(punc_id) if punc_id is not None else 1
+ sentence_end = time_stamp[1] if time_stamp is not None else sentence_end
+
+ if punc_id > 1:
+ sentence_text += punc_list[punc_id - 2]
res.append({
'text': sentence_text,
"start": sentence_start,
- "end": time_stamp_postprocessed[i][1]
+ "end": sentence_end,
+ "text_seg": sentence_text_seg,
+ "ts_list": ts_list
})
sentence_text = ''
- sentence_start = time_stamp_postprocessed[i][1]
- elif punc_id_list[i] == 3:
- sentence_text += '.'
- res.append({
- 'text': sentence_text,
- "start": sentence_start,
- "end": time_stamp_postprocessed[i][1]
- })
- sentence_text = ''
- sentence_start = time_stamp_postprocessed[i][1]
+ sentence_text_seg = ''
+ ts_list = []
+ sentence_start = sentence_end
return res
@@ -134,9 +169,9 @@
uttid_intersection = self._intersection(uttid_list1, uttid_list2)
res = self.as_cal(uttid_intersection, ts_dict1, ts_dict2)
logging.warning("Average shift of {} and {}: {}.".format(file1, file2, str(res)[:8]))
- logging.warning("Following timestamp pair differs most: {}, detail:{}".format(self.max_shift_uttid))
+ logging.warning("Following timestamp pair differs most: {}, detail:{}".format(self.max_shift, self.max_shift_uttid))
- def _intersection(list1, list2):
+ def _intersection(self, list1, list2):
set1 = set(list1)
set2 = set(list2)
if set1 == set2:
@@ -170,8 +205,9 @@
return uttid_list, ts_dict
def _shift(self, filtered_timestamp_list1, filtered_timestamp_list2):
+ shift_time = 0
for fts1, fts2 in zip(filtered_timestamp_list1, filtered_timestamp_list2):
- shift_time = abs(fts1[0] - fts2[0]) + abs(fts1[1] - fts2[1])
+ shift_time += abs(fts1[0] - fts2[0]) + abs(fts1[1] - fts2[1])
num_tokens = len(filtered_timestamp_list1)
return shift_time, num_tokens
@@ -232,13 +268,54 @@
return self._accumlated_shift / self._accumlated_tokens
-SUPPORTED_MODES = ['cal_aas']
+def convert_external_alphas(alphas_file, text_file, output_file):
+ from funasr.models.predictor.cif import cif_wo_hidden
+ with open(alphas_file, 'r') as f1, open(text_file, 'r') as f2, open(output_file, 'w') as f3:
+ for line1, line2 in zip(f1.readlines(), f2.readlines()):
+ line1 = line1.rstrip()
+ line2 = line2.rstrip()
+ assert line1.split()[0] == line2.split()[0]
+ uttid = line1.split()[0]
+ alphas = [float(i) for i in line1.split()[1:]]
+ new_alphas = np.array(remove_chunk_padding(alphas))
+ new_alphas[-1] += 1e-4
+ text = line2.split()[1:]
+ if len(text) + 1 != int(new_alphas.sum()):
+ # force resize
+ new_alphas *= (len(text) + 1) / int(new_alphas.sum())
+ peaks = cif_wo_hidden(torch.Tensor(new_alphas).unsqueeze(0), 1.0-1e-4)
+ if " " in text:
+ text = text.split()
+ else:
+ text = [i for i in text]
+ res_str, _ = ts_prediction_lfr6_standard(new_alphas, peaks[0], text,
+ force_time_shift=-7.0,
+ sil_in_str=False)
+ f3.write("{} {}\n".format(uttid, res_str))
+
+
+def remove_chunk_padding(alphas):
+ # remove the padding part in alphas if using chunk paraformer for GPU
+ START_ZERO = 45
+ MID_ZERO = 75
+ REAL_FRAMES = 360 # for chunk based encoder 10-120-10 and fsmn padding 5
+ alphas = alphas[START_ZERO:] # remove the padding at beginning
+ new_alphas = []
+ while True:
+ new_alphas = new_alphas + alphas[:REAL_FRAMES]
+ alphas = alphas[REAL_FRAMES+MID_ZERO:]
+ if len(alphas) < REAL_FRAMES: break
+ return new_alphas
+
+SUPPORTED_MODES = ['cal_aas', 'read_ext_alphas']
def main(args):
if args.mode == 'cal_aas':
asc = AverageShiftCalculator()
asc(args.input, args.input2)
+ elif args.mode == 'read_ext_alphas':
+ convert_external_alphas(args.input, args.input2, args.output)
else:
logging.error("Mode {} not in SUPPORTED_MODES: {}.".format(args.mode, SUPPORTED_MODES))
--
Gitblit v1.9.1