Abstract | ||
---|---|---|
We present a text-to-speech (TTS) system designed for the dialect of Bengali spoken in Bangladesh. This work is part of an ongoing effort to address the needs of new under-resourced languages. We propose a process for streamlining the bootstrapping of TTS systems for under-resourced languages. First, we use crowdsourcing to collect the data from multiple ordinary speakers, each speaker recording small amount of sentences. Second, we leverage an existing text normalization system for a related language (Hindi) to bootstrap a linguistic front-end for Bangla. Third, we employ statistical techniques to construct multi-speaker acoustic models using Long Short-term Memory Recurrent Neural Network (LSTM-RNN) and Hidden Markov Model (HMM) approaches. We then describe our experiments that show that the resulting TTS voices score well in terms of their perceived quality as measured by Mean Opinion Score (MOS) evaluations. |
Year | DOI | Venue |
---|---|---|
2016 | 10.1016/j.procs.2016.04.049 | Procedia Computer Science |
Keywords | Field | DocType |
TTS,Bangladesh,HMM,LSTM-RNN,acoustic modeling | Crowdsourcing,Hindi,Computer science,Bootstrapping,Recurrent neural network,Speech recognition,Mean opinion score,Bengali,Artificial intelligence,Natural language processing,Hidden Markov model,Text normalization | Conference |
Volume | ISSN | Citations |
81 | 1877-0509 | 2 |
PageRank | References | Authors |
0.40 | 15 | 6 |
Name | Order | Citations | PageRank |
---|---|---|---|
Alexander Gutkin | 1 | 2 | 1.08 |
Linne Ha | 2 | 5 | 3.19 |
Martin Jansche | 3 | 257 | 23.92 |
Oddur Kjartansson | 4 | 6 | 4.89 |
Knot Pipatsrisawat | 5 | 358 | 20.44 |
Richard Sproat | 6 | 31 | 7.34 |