Intra- and Inter-frame Features for Automatic Speech Recognition

Abstract

In this paper, alternative dynamic features for speech recognition are proposed. The goal of this work is to improve speech recognition accuracy by deriving the representation of distinctive dynamic characteristics from a speech spectrum. This work was inspired by two temporal dynamics of a speech signal. One is the highly non‐stationary nature of speech, and the other is the inter‐frame change of a speech spectrum. We adopt the use of a sub‐frame spectrum analyzer to capture very rapid spectral changes within a speech analysis frame. In addition, we attempt to measure spectral fluctuations of a more complex manner as opposed to traditional dynamic features such as delta or double‐delta. To evaluate the proposed features, speech recognition tests over smartphone environments were conducted. The experimental results show that the feature streams simply combined with the proposed features are effective for an improvement in the recognition accuracy of a hidden Markov model–based speech recognizer.

Natural Language Dialog Systems and Intelligent Assistants

“…We first extracted speech feature vectors at every 10 ms for a 20 ms analysis window Lee et al 2014). Then, we trained separately native acoustic models (AMs) and non-native AMs by using native and non-native utterances, respectively (Young et al 2009).…”

Weighted Finite State Transducer-Based Endpoint Detection Using Probabilistic Decision Logic

“…1 shows widely used decision logic [2]. This condition is usually based on heuristic knowledge, and Fig.…”

“…where F is a frame-level WFST representing equations (1) and (2), and where U is an utterance-level WFST representing equation (4) with two additional output symbols to mark the begin-of-utterance (BOU) and end-of-utterance (EOU). The endpoint is detected if the output symbol of the last transition of the best path P, o(e t ), satisfies the following condition:…”

“…However, most research activities have focused on improving the frame-level decision performance by developing robust features [1]- [2], feature combinations [3], and modeling approaches [4]- [9], while little attention has been paid to utterance-level decision, integrating both decision processes, or improving both processes at the same time. The first process is frame-level speech/non-speech classification based on statistical hypothesis testing, and the second process involves utterance-level speech boundary decision based on heuristic knowledge.…”

Online Blind Channel Normalization Using BPF-Based Modulation Frequency Filtering

“…After filtering with the proposed blind normalization filtering approach, 13 MFCCs were extracted by taking a discrete cosine transform. We then derived 39 dynamic feature vectors (inter-frame features) and one intra-log energy measure from the 13 MFCC features [2]. For our speech recognition experiments, we used 53 feature vector sequences (13 MFCCs + 39 dynamic features + 1 intra-log energy).…”