zev/tpro
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

finished refactoring to a single repo, and to OOP for straight-forward adding of new ASR APIs. added Gentle, and added viral_overlay JSON output. added tests

Browse Source
This commit is contained in:
zevav
2019-02-06 22:28:08 -05:00
parent 84fe4d2fd4
commit e36c8ba30e
11 changed files with 347 additions and 103 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
converters/__init__.py
View File

@@ -1,24 +0,0 @@
"""
fields for converted transcript:
start
end
word
confidence
index
always_capitalized
punc_before
punc_after
"""
from transcript_processing.converters.amazon import amazon_converter
from transcript_processing.converters.speechmatics import speechmatics_aligned_text_converter, speechmatics_converter
converters = {
'speechmatics': speechmatics_converter,
'speechmatics_align': speechmatics_aligned_text_converter,
'amazon': amazon_converter,
}

76
converters/amazon.py
View File

@@ -1,6 +1,7 @@
import json
from transcript_processing import helpers
from converter import TranscriptConverter
import helpers
@@ -10,11 +11,11 @@ class AmazonConverter(TranscriptConverter):
super().__init__(path, output_target)
def get_word_objects(self, json_data):
return data['results']['items']
return json_data['results']['items']
def get_words(self, word_objects):
return [self.get_word_word(w)
for w in word_objects])
for w in word_objects]
@staticmethod
def get_word_start(word_object):
@@ -30,7 +31,7 @@ class AmazonConverter(TranscriptConverter):
@staticmethod
def get_word_word(word_object):
word_word = w['alternatives'][0]['content']
word_word = word_object['alternatives'][0]['content']
if word_word == 'i':
# weird Amazon quirk
word_word = 'I'
@@ -44,11 +45,11 @@ class AmazonConverter(TranscriptConverter):
num_words = len(words)
index = 0
for i, w in enumerate(words):
for i, w in enumerate(word_objects):
if w['type'] == 'punctuation':
continue
next_word_punc_after = None
word_obj = self.get_word_object(w, i, tagged_words, words)
word_obj = self.get_word_object(w, i, tagged_words, word_objects)
if word_obj.next_word:
next_word = self.get_word_word(word_obj.next_word)
@@ -60,7 +61,7 @@ class AmazonConverter(TranscriptConverter):
next_word_punc_after = None
if word_obj.word.lower() == 'you' and next_word == 'know':
prev_word = words[i - 1]
prev_word = word_objects[i - 1]
if prev_word['type'] != 'punctuation':
converted_words[-1]['punc_after'] = ','
if next_word_type != 'punctuation':
@@ -83,64 +84,3 @@ class AmazonConverter(TranscriptConverter):
punc_after = False
return converted_words
def amazon_converter(data: dict):
data = json.load(data)
converted_words = []
words = data['results']['items']
tagged_words = helpers.tag_words(
[w['alternatives'][0]['content'] for w in words])
punc_before = False
punc_after = False
num_words = len(words)
index = 0
for i, w in enumerate(words):
if w['type'] == 'punctuation':
continue
next_word_punc_after = None
word_start = float(w['start_time'])
word_end = float(w['end_time'])
confidence = float(w['alternatives'][0]['confidence'])
word = w['alternatives'][0]['content']
is_proper_noun = tagged_words[i][1] in helpers.PROPER_NOUN_TAGS
next_word = None
if i < num_words - 1:
next_word = words[i + 1]['alternatives'][0]['content']
next_word_type = words[i + 1]['type']
if next_word == '.':
punc_after = '.'
elif next_word == ',':
punc_after = ','
elif next_word_punc_after:
punc_after = next_word_punc_after
next_word_punc_after = None
if word == 'i':
# weird Amazon quirk
word = 'I'
if word.lower() == 'you' and next_word == 'know':
prev_word = words[i - 1]
if prev_word['type'] != 'punctuation':
converted_words[-1]['punc_after'] = ','
if next_word_type != 'punctuation':
next_word_punc_after = ','
converted_words.append({
'start': word_start,
'end': word_end,
'confidence': confidence,
'word': word,
'always_capitalized': is_proper_noun or word == 'I',
'index': index,
'punc_after': punc_after,
'punc_before': punc_before,
})
index += 1
punc_after = False
return converted_words

60
converters/gentle.py Normal file
View File

@@ -0,0 +1,60 @@
from converter import TranscriptConverter
class GentleConverter(TranscriptConverter):
def __init__(self, path, output_target):
super().__init__(path, output_target)
def get_word_objects(self, json_data):
return json_data['words']
def get_words(self, word_objects):
return [self.get_word_word(w)
for w in word_objects]
@staticmethod
def get_word_start(word_object):
return word_object['start']
@staticmethod
def get_word_end(word_object):
return word_object['end']
@staticmethod
def get_word_confidence(word_object):
return 1
@staticmethod
def get_word_word(word_object):
return word_object['alignedWord']
def convert_words(self, word_objects, words, tagged_words=None):
converted_words = []
punc_before = False
punc_after = False
num_words = len(words)
index = 0
for i, w in enumerate(word_objects):
word_obj = self.get_word_object(w, i, tagged_words, word_objects)
converted_words.append({
'start': word_obj.start,
'end': word_obj.end,
'confidence': word_obj.confidence,
'word': word_obj.word,
'always_capitalized': (
word_obj.is_proper_noun
or word_obj.word == 'I'),
'index': index,
'punc_after': punc_after,
'punc_before': punc_before,
})
index += 1
punc_after = False
return converted_words

73
converters/speechmatics.py
View File

@@ -1,10 +1,74 @@
from collections import namedtuple
import json
from transcript_processing import helpers
from converter import TranscriptConverter
import helpers
Word = namedtuple('Word', 'start end word')
class SpeechmaticsConverter(TranscriptConverter):
def __init__(self, path, output_target):
super().__init__(path, output_target)
def get_word_objects(self, json_data):
return json_data['words']
def get_words(self, word_objects):
return [self.get_word_word(w)
for w in word_objects]
@staticmethod
def get_word_start(word_object):
return float(word_object['time'])
@staticmethod
def get_word_end(word_object):
return (SpeechmaticsConverter.get_word_start(word_object)
+ float(word_object['duration']))
@staticmethod
def get_word_confidence(word_object):
return float(word_object['confidence'])
@staticmethod
def get_word_word(word_object):
return word_object['name']
def convert_words(self, word_objects, words, tagged_words=None):
converted_words = []
punc_before = False
punc_after = False
num_words = len(words)
index = 0
for i, w in enumerate(word_objects):
word_obj = self.get_word_object(w, i, tagged_words, word_objects)
if word_obj.word == '.':
continue
if word_obj.next_word:
next_word = self.get_word_word(word_obj.next_word)
if next_word == '.':
punc_after = '.'
converted_words.append({
'start': word_obj.start,
'end': word_obj.end,
'confidence': word_obj.confidence,
'word': word_obj.word,
'always_capitalized': (
word_obj.is_proper_noun
or word_obj.word == 'I'),
'index': index,
'punc_after': punc_after,
'punc_before': punc_before,
})
index += 1
punc_after = False
return converted_words
def speechmatics_converter(data):
@@ -55,6 +119,8 @@ def speechmatics_aligned_text_converter(data):
class Exhausted(Exception):
pass
Word = namedtuple('Word', 'start end word')
def get_time(transcript, index):
time_index = transcript.find('time=', index)
if time_index == -1:
@@ -108,6 +174,3 @@ def speechmatics_aligned_text_converter(data):
})
return converted_words
def gentle_converter