| Presentation | 2001/9/21 Separation of concurrent sounds based on the excitation sources using the time-domain : Gain-adapted AR-HMM decomposition method Akira SASOU, Kazuyo TANAKA, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | This paper describes a model-based method of separating monaural concurrent sounds. The proposed method adopts a source filter model that consists of an Auto-Regressive process for an articulatory filter and a Hidden Markov Model for an excitation source. In addition, the source filter model represents waveform amplitude variations by multiplying the HMM's output by a time-varying gain. The proposed method achieves sound separation based on estimates of the state transition sequence of HMM and the time series of gain for the each AR-HMM from mixed sounds. We refer to the proposed method as the gain-adapted AR-HMM decomposition method. The experiment was conducted by applying the method to a separation of two amplitude-modulated vowel sounds. The experiment results show that the proposed method can separate each sound's waveform with an SNR of approximately 9[dB], irrespective of the modulation index. In an experiment applying the proposed method to separate vowel sound and white Gaussian noise with time variance, the noise-degraded vowel sound at SNR 3.6[dB]was enhanced to 9.2[dB]. These experimental results confirmed that the proposed method has the ability to separate mixtures of periodic and general aperiodic sounds. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | monaural concurrent sounds / sound-separation / noise reduction / source filter / AR-HMM |
| Paper # | DSP2001-96,SP2001-69 |
| Date of Issue |
| Conference Information | |
| Committee | SP |
|---|---|
| Conference Date | 2001/9/21(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Speech (SP) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Separation of concurrent sounds based on the excitation sources using the time-domain : Gain-adapted AR-HMM decomposition method |
| Sub Title (in English) | |
| Keyword(1) | monaural concurrent sounds |
| Keyword(2) | sound-separation |
| Keyword(3) | noise reduction |
| Keyword(4) | source filter |
| Keyword(5) | AR-HMM |
| 1st Author's Name | Akira SASOU |
| 1st Author's Affiliation | National Institute of Advanced Industrial Science and Technology() |
| 2nd Author's Name | Kazuyo TANAKA |
| 2nd Author's Affiliation | National Institute of Advanced Industrial Science and Technology |
| Date | 2001/9/21 |
| Paper # | DSP2001-96,SP2001-69 |
| Volume (vol) | vol.101 |
| Number (no) | 325 |
| Page | pp.pp.- |
| #Pages | 8 |
| Date of Issue | |