Subphonetic Modeling for Speech Recognition

Subphonetic Modeling for Speech Recognition Mei-Yuh Hwang Xuedong Huang School of Computer Science Carnegie Mellon University Pittsburgh, Pennsylvania 15213 ABSTRACT How to capture important acoustic clues and estimate essen- tial parameters reliably is one of the central issues in speech recognition, since we will never have sufficient training data to model various acoustic-phonetic phenomena. Success- ful examples include subword models with many smoothing techniques. In comparison with subword models, subphonetic modeling may provide a finer level of details. We propose to model subphonetic events with Markov states and treat the state in phonetic hidden Markov models as our basic sub- phonetic unit -- senone. A word model is a concatenation of state-dependent senones and senones can be shared across different word models. Senones not only allow parameter sharing, but also enable pronunciation optimization and new word learning, where the phonetic baseform is replaced by the senonic baseform. In this paper, we report preliminary subphonetic modeling results, which not only significantly re- duced the word error rate for speaker-independent continuous speech recognition but also demonstrated a novel application for new word learning. 1 INTRODUCTION For large-vocabulary speech recognition, we will never have sufficient training data to model all the various acoustic- phonetic phenomena. How to capture important acoustic clues and estimate essential parameters reliably is one of the central issues in speech recognition. To share parame- ters among different word modes, context-dependent subword models have been used successfully in many state-of-the-art speech recognition systems [1, 2, 3, 4]. The principle of pa- rameter sharing can also be extended to subphonetic models. For subphonetic modeling, fenones [5, 6] have been used as the front end output of the IBM acoustic processor. To gener- ate a fenonic pronunciation, multiple examples of each word are obtained. The fenonic baseform is built by searching for a sequence of fenones which has the maximum probability of generating all the given multiple utterances. The codeword- dependent fenonic models are then trained just like phonetic models. We believe that the 200 codeword-dependent fenones may be insufficient for large-vocabulary continuous speech recognition. In this paper, we propose to model subphonetic events with Markov states. We will treat the state in hidden Markov models (HMMs) as a basic subphonetic unit -- senone. The total number of HMM states in a system is often too large to be well trained. To reduce the number of free parame- ters, we can cluster the state-dependent output distributions. Each clustered output distribution is denoted as a senone. In this way, senones can be shared across different models as illustrated in Figure 1. senone codebook Figure 1: Creation of the senone codebook. The advantages of senones include better parameter sharing and improved pronunciation optimization. After clustering, different states in different models may share the same senone if they exhibit acoustic similarity. Clustefingat the granularity of the state rather than the entire model (like generalized tripbones) can keep the dissimilar states of two similar models apart while the other corresponding states are merged, and thus lead …

Content truncated to 500 words