Simon/Base models: Difference between revisions
m (moved Simon: Base models to Simon/Base models) |
No edit summary |
||
| Line 46: | Line 46: | ||
* A [http://www.laps.ufpa.br/falabrasil/downloads.php brazilian portuguese acoustic model] made by [http://www.laps.ufpa.br/falabrasil/ the FalaBrasil project] especially for Simon ([http://www.laps.ufpa.br/falabrasil/files/LapsAM-1.5-16k-SimonBR.rar Direct Link]). | * A [http://www.laps.ufpa.br/falabrasil/downloads.php brazilian portuguese acoustic model] made by [http://www.laps.ufpa.br/falabrasil/ the FalaBrasil project] especially for Simon ([http://www.laps.ufpa.br/falabrasil/files/LapsAM-1.5-16k-SimonBR.rar Direct Link]). | ||
[[Category:Accessibility]] | |||
Revision as of 18:01, 17 August 2012
Base models are a way of using existing (often speaker independant) acoustic models with Simon.
Create a speech model
To create your own speech model, you need to install the HTK.
With Simon you can easily create your own personal speech model. You don't need any extra files for this.
Static base model
If you are creating solutions which will be used by more than one user or simply don't have the time to train the system you can use static base models.
Static models are used as-is and are not modified by Simon in any way.
Because of this it is important that the selected base model matches your voice as closely as possible.
Adapt base model
To create your own speech model, you need to install the HTK.
You can select to adapt your static base models (see above). In this case, Simon will adapt the generic base model to your voice during model generation. This will improve the recognition rate of the resulting model.
If you choose to use adapted base models but don't record any trainingssamples, Simon will use the base models to be adapted just like static models.
Where to get base models
This depends on your target language.
To import the target model, download the archive and extract the following files: hmmdefs, tiedlist, macros and stats. To use the model just configure Simon to your wishes (Settings > Configure Simon > Model settings) and point Simon to those four files of the downloaded base model.
Your speech model will be automatically generated out of your trainings samples.
English
- Voxforge; You can find their models in the download section. You need to look for the nightly built 16kHz, 16bit models. Direct link to the nightly builds
- HTK Wall Street Journal; Untested by me but should work fine. Please edit this if you confirm that this is working.
German
- Voxforge; You can find their models in the download section. You need to look for Simon models. Direct link
Portuguese
- A brazilian portuguese acoustic model made by the FalaBrasil project especially for Simon (Direct Link).