Click here to register.

Edit Message

Visitor Name
Subject
Message
Attachment

Re: Problems setting an 8kHz model on Sphinx 4

From the Sphinx-4 FAQ: Can I use Sphinx-4 to recognize telephone audio?

Can I use Sphinx-4 to recognize telephone audio?

    The issue with telephone audio is that it has limited range of frequences. Unlike usual microphone recording that includes frequences from 1 Hz to 8000 kHz, telephone audio is passed through frequency filters. As a result telephone audio contains frequences from 200 Hz to 3500 Hz. That makes it impossible to recognize telephone audio with usual microphone acoustic model. You need to use specialized models to recognize it.

    There are few common telephone models distributed which you can use. Most notably, Communicator models, WSJ_8k model from sphinx4 and Voxforge English model.

    To configure sphinx4 with 8kHz model change two things mel filter parameters and model itself:

      <component name="melFilterBank" type="edu.cmu.sphinx.frontend.frequencywarp.MelFrequencyFilterBank">
        <property name="numberFilters" value="31"/>
        <property name="minimumFrequency" value="200"/>
        <property name="maximumFrequency" value="3500"/>
      </component>

      <component name="sphinx3Loader"
                type="edu.cmu.sphinx.linguist.acoustic.tiedstate.Sphinx3Loader">
        <property name="logMath" value="logMath"/>
        <property name="unitManager" value="unitManager"/>
        <property name="location" value="the path to the model folder"/>
      </component>

      <component name="wsjLoader" type="edu.cmu.sphinx.linguist.acoustic.tiedstate.Sphinx3Loader">
        <property name="logMath" value="logMath"/>
        <property name="unitManager" value="unitManager"/>
        <property name="location" value="resource:/WSJ_8gau_13dCep_8kHz_31mel_200Hz_3500Hz"/>
        <property name="modelDefinition" value="etc/WSJ_8gau_13dCep_8kHz_31mel_200Hz_3500Hz.4000.mdef"/>
        <property name="dataLocation" value="cd_continuous_8gau/"/>
      </component>