In this work, we automatically identify and disambiguate homophones in code-switched data to improve recognition of code-switched speech. We also extend this framework to propose a metric for code-switched speech recognition that takes into account homophones in both languages while calculating WER.