An HMM/DNN framework is proposed to address the issues of short-word detection. The first-stage keyword hypothesizer is redesigned with a context-aware keyword model and a 9state filler model to reduce the miss rate from 80% to 6% and increase the figure-of-merit (FOM) from 6.08% to 21.88% for short words. The hypothesizer is followed by a MLP-based second-stage keyword verifier to further reduce its putative hits. To enhance short word detection, three new techniques, including an HMM-based feature transformation for the MLPs, knowledge-based features, and deep neural networks, are incorporated into redesigning the verifier. With a set of nine short keywords from the TIMIT set the best FOM we had achieved for the proposed KWS system was 42.79%, which is comparable with that of 42.6% for long content words and much better than the FOM of 18.4% for short keywords reported in previous research [10].

A Hybrid HMM/DNN Approach to Keyword Spotting of Short Words

https://www.cstr.ed.ac.uk/downloads/publications/2013/Tejedor_CSL2013.pdf

Feature analysis for discriminative confidence estimation in spoken term detection

A hierarchical framework is proposed to address the issues of modeling different type of words in keyword spotting (KWS). Keyword models are built at various levels according to the availability of training set resources for each individual word. The proposed approach improves the performance of KWS even when no training speech is available for the keywords. It also suggests an easier way to collect training data for these resource-limited words. Experimental results show that the proposed framework improves performance in KWS in a figure-of-merit (FOM) metric regardless of the number of training instances for each keyword. For words with abundant speech data, the proposed method exploits the training data better than the conventional modeling technique and boosts the system FOM from 9.79% to 42.78%. For words with a small amount of training data, the new method increases the system FOM from 29.05% to 49.06%. Even for keywords without any training examples, the new modeling scheme improves the system FOM from 60.96% to 66.51%.

A resource-dependent approach to word modeling for keyword spotting.

A keyword-sensitive language modeling framework for spoken keyword search (KWS) is proposed to combine the advantages of conventional keyword-filler based and large vocabulary continuous speech recognition (LVCSR) based KWS systems. The proposed framework allows keyword search systems to be flexible on keyword target settings as in the LVCSR-based keyword search. In low-resource scenarios it facilitates KWS with an ability to achieve high keyword detection accuracy as in the keyword-filler based systems and to attain a low false alarm rate inherent in the LVCSR-based systems. The proposed keyword-aware grammar is realized by incorporating keyword information to re-train and modify the language models used in LVCSR-based KWS. Experimental results, on the evalpart1 data of the IARPA Babel OpenKWS13 Vietnamese tasks, indicate that the proposed approach achieves a relative improvement, over the conventional LVCSR-based KWS systems, of the actual term weighted value for about 57 % (from 0.2093 to 0.3287) and 20 % (from 0.4578 to 0.5486) on the limited-language-pack and full-language-pack tasks, respectively.

A Keyword-Aware Language Modeling Approach to Spoken Keyword Search

Keyword spotting is a process of finding some specific words or phrases in recorded speeches by computers. Deep neural network algorithms, as a powerful engine, can handle this problem if they are trained over an appropriate dataset. To this end, the football keyword dataset (FKD), as a new keyword spotting dataset in Persian, is collected with crowdsourcing. This dataset contains nearly 31000 samples in 18 classes. The continuous speech synthesis method proposed to made FKD usable in the practical application which works with continuous speeches. Besides, we proposed a lightweight architecture called EfficientNet-A0 (absolute zero) by applying the compound scaling method on EfficientNet-B0 for keyword spotting task. Finally, the proposed architecture is evaluated with various models. It is realized that EfficientNet-A0 and Resnet models outperform other models on this dataset.

EfficientNet-Absolute Zero for Continuous Speech Keyword Spotting.

We focus on utterance verification of short keywords, which contain only one syllable. The relative less information makes the task rather tough. We integrate neural network and hidden Markov models (HMMs) in an attempt to utilize the strength of both. A classifier was built for each keyword. Our model differs from other hybrid models in that we feed the neural networks with posterior likelihood generated by a set of HMMs. Therefore, statistical information can be reserved and the size of neural networks is not too large. Another feature is our utilization of word duration. Our experiments were carried out in Mandarin. Two baseline models for comparison were built, one employed the likelihood ratio measure and the other utilized a neural network presented in preceding studies. Two models depending on our approaches were constructed. The results showed that one proposed approach improved by 8.7% and 12.5%, and another by 15.2% and 19.2%.

http://ieeexplore.ieee.org/iel5/7719/21186/00983657.pdf

Utterance verification of short keywords using hybrid neural-network/HMM approach

Speaker adaptation in transcription-mode speech recognition means adaptation on the test set itself rather than on an independent data set. A confidence-driven iterative adaptation approach is presented. Speech recognition and speaker adaptation are performed alternately. A combination of MLLR (maximum likelihood linear regression) and MAP (maximum a posteriori) approaches is used for adaptation. To avoid erroneous transcription in adaptation, an N-best based confidence measure is introduced. The procedure includes a coarse classification and a delicate measurement. Experiments were carried out on a large vocabulary Mandarin continuous speech recognition. A baseline model for comparison was built. The experimental results showed that our proposed approach reduced the word error rate by 51.8% compared to the initial speaker-independent system and 39.3% compared to the baseline model.

Kaijiang Chen

Papers

Utterance verification of short keywords using hybrid neural-network/HMM approach

Confidence-driven iterative speaker adaptation in transcription-mode speech recognition