Merge pull request #492 from alibaba-damo-academy/dev_smohan

yhliang

2023-05-11 d2dc3af1a69ee4075bcfc0c83dc0fb8e3fc1db4e

 funasr/runtime/python/onnxruntime/funasr_onnx/punc_bin.py

@@ -1,4 +1,6 @@
# -*- encoding: utf-8 -*-
# Copyright FunASR (https://github.com/alibaba-damo-academy/FunASR). All Rights Reserved.
#  MIT License  (https://opensource.org/licenses/MIT)

import os.path
from pathlib import Path
@@ -13,6 +15,11 @@


class CT_Transformer():
    """
    Author: Speech Lab of DAMO Academy, Alibaba Group
    CT-Transformer: Controllable time-delay transformer for real-time punctuation prediction and disfluency detection
    https://arxiv.org/pdf/2003.01309.pdf
    """
    def __init__(self, model_dir: Union[str, Path] = None,
                 batch_size: int = 1,
                 device_id: Union[str, int] = "-1",
@@ -57,7 +64,7 @@
            mini_sentence = mini_sentences[mini_sentence_i]
            mini_sentence_id = mini_sentences_id[mini_sentence_i]
            mini_sentence = cache_sent + mini_sentence
            mini_sentence_id = np.array(cache_sent_id + mini_sentence_id, dtype='int64')
            mini_sentence_id = np.array(cache_sent_id + mini_sentence_id, dtype='int32')
            data = {
                "text": mini_sentence_id[None,:],
                "text_lengths": np.array([len(mini_sentence_id)], dtype='int32'),
@@ -119,6 +126,11 @@


class CT_Transformer_VadRealtime(CT_Transformer):
    """
    Author: Speech Lab of DAMO Academy, Alibaba Group
    CT-Transformer: Controllable time-delay transformer for real-time punctuation prediction and disfluency detection
    https://arxiv.org/pdf/2003.01309.pdf
    """
    def __init__(self, model_dir: Union[str, Path] = None,
                 batch_size: int = 1,
                 device_id: Union[str, int] = "-1",
@@ -136,7 +148,7 @@
        else:
            precache = ""
            cache = []
        full_text = precache + text
        full_text = precache + " " + text
        split_text = code_mix_split_words(full_text)
        split_text_id = self.converter.tokens2ids(split_text)
        mini_sentences = split_to_mini_sentence(split_text, split_size)
@@ -154,7 +166,7 @@
            mini_sentence = mini_sentences[mini_sentence_i]
            mini_sentence_id = mini_sentences_id[mini_sentence_i]
            mini_sentence = cache_sent + mini_sentence
            mini_sentence_id = np.concatenate((cache_sent_id, mini_sentence_id), axis=0)
            mini_sentence_id = np.concatenate((cache_sent_id, mini_sentence_id), axis=0,dtype='int32')
            text_length = len(mini_sentence_id)
            data = {
                "input": mini_sentence_id[None,:],