Manuscript speech is a common type of speech in various official events. However, we often observe that many speakers simply read the script without making eye contact, thereby lowering audience engagement. Practice with proper tools could benefit a speaker considerably. In this work, we iteratively designed ScriptFree, an adaptive speech practice environment where off-the-shelf automatic speech recognition (ASR) is leveraged to measure a speaker's preparation level, and accordingly, the script is adaptively compressed to reduce the speaker's visual reliance toward script mastery. The user study results confirmed that ScriptFree helped the participants to success- fully improve their speech over multiple practice iterations. The results have significant design implications for building adaptive speech practice systems.