Yes. See: "VTLN". Also, the audio feature transformations are adapted during recognition in most cases. For instance, speaker independent recognizers generally do runtime adaptation on a per conversation basis. Speaker dependent ones generally continually adapt to the user using the same techniques.