...
Overview
The Google and AT&T Automatic Speech Recognition (ASR) is a java application which acts as a speech server which uses either Google's speech recognition engine or AT&T's speech recognition engine based on what you select. Both Google and AT&T servers, implementations share the same code (it's a jar that implements the sonic protocol and both use the same jar).
They are both web services and so depend on having an Internet therefore require an internet connection.
Both services do not provide partial interpretations in real time, just the final interpretation. To use them one needs acquirespeechThis component is used in connection with Acquirespeech. The audio is accumulated while the push-to-talk is pressed and sent it over to the server when the button is released. The results are then received and communicated back to acquirespeechAcquirespeech.
Quick facts:
- Location: core/GoogleASR
- Language: Java
- Distribution: source
...
The AT&T service requires a subscription and allows for custom language models to be trained. Once you subscribe, you will receive a username and password to access your account.
You can find more details on how to subscribe and use the ASR engine here.
In order to create a language model
- prepare a file with one utterance ,that you want to be recognized, for each line of the file
- then select
...
- 'Manage grammar files'
- then select upload and select the text file (it needs to have the .train extension)
- then compile it
- then you can rename it to
...
- any other id you want to use)
Note: In order to use the AT&T ASR from the Toolkit launcher. Select "ATNT" as your speech server and your client type. you will need to provide your username and password as command line parameters. This can be done in the space provided for it in the launcher as shown below.
Using the Google ASR
The Google ASR doesn't require anything and cannot be customized (but it's very good in its the current settings work pretty well). the The major limitation of the Google ASR is that it processes audios only about audio with a max length of 10 seconds long.
Command line parameters
...
usage: [-p TCP port ][-m seconds ] [-h] [-t G|A (G is default)] [-u AT&T_MashUp_UID ] [-l Language model ID for AT&T ]
Known Issues
- List of common known issues, like why something isn't working, why it's implemented in a certain way, limitations, etc. If there are major Jira tickets, link to those as well. Google ASR only processes 10 second maximum for the audio.
FAQ
See the Main FAQ. Please use the Google Groups emailing list for unlisted questions.