I know teams which has their sprint length floating so that's not an issue to change sprint length over time.
The main value you get from keeping the sprints equally-sized is you can measure how much work the team can do over the next chunk of time. If you change sprint length during the project it will be more difficult to estimate velocity.
Besides in Scrum sprint length determines frequency of planning meetings, demos and retrospectives. If the team is new to Scrum I'd probably go for shorter iterations from the beginning - you don't have to treat as the rule that you have to deliver presentable demo after each sprint, but the more frequently you meet to improve your process (retrospective!) the faster you'll be adjusting your method to the optimal one.
In short: if you have rather experienced team and measuring velocity isn't crucial I wouldn't object changing sprint during the project. However if the team isn't that experienced (so you prepare for frequent process adjustments) or learning how fast you can run is important I'd prefer to start with 2-week-long sprints on the beginning.