Loading...
Please wait, while we are loading the content...
Similar Documents
Using intonation to constrain language models in speech recognition (1997)
| Content Provider | CiteSeerX |
|---|---|
| Author | Malinowski, Simon Jégou, Hervé Guillemot, Christine |
| Abstract | Methodologies to analyze error recovery properties of Variable Length Codes (VLCs) have been introduced in [1] and [2]. In this paper, we extend these methods to analyze the error-resilience of VLCs when soft decoding with length constraint strategies are applied at the decoder side. The approach allows in particular to compute the amount of information conveyed by the length constraint on a trellis, and hence the maximum amount of information that a soft VLC decoder, augmented with a length constraint, will be able to exploit. Then, this amount of information, as well as the probability that the VLC decoder does not re-synchronize in a strict sense, are shown not to be significantly altered by appropriate trellis states aggregation. This proves that the performances of a Viterbi decoder run on aggregated state models with a length constraint can be optimal with a significantly reduced complexity compared with the bit/symbol trellis. 1 |
| File Format | |
| Volume Number | 2763 |
| Journal | EUROSPEECH-97 |
| Language | English |
| Publisher Date | 1997-01-01 |
| Access Restriction | Open |
| Subject Keyword | Speech Recognition Language Model Length Constraint Length Constraint Strategy Error Recovery Property Soft Vlc Decoder Vlc Decoder Variable Length Code Soft Decoding Decoder Side Appropriate Trellis State Aggregation Aggregated State Model Viterbi Decoder Run Maximum Amount Bit Symbol Trellis Strict Sense |
| Content Type | Text |
| Resource Type | Article |
