This study developed and tested a novel automated method for applying verbal self-feedback for training patients over multiple sessions. Verbal self-feedback is a novel approach to treatment that uses recursive post-production feedback to fine-tune speech production. This approach is based on studies on the facilitatory role of recursive auditory feedback in speech production and vocal learning in humans and songbirds respectively. We used a cross-over design to compare the effect of verbal self-feedback (experimental treatment) to that of script training (control treatment) on speech fluency in two patients with chronic mild nonfluent aphasia (AE2 and AE3). We developed a smartphone App for feedback training and measured speech fluency in two participants that received both experimental and control treatments sequentially, in counterbalanced order. Each treatment comprised two-hour daily sessions over three weeks, with two weeks washout between treatments. Both treatments used scripts, each comprising eight sentences. Direct treatment effects were measured by comparing speech fluency scores of the first day of treatment with the last day of treatment on trained scripts. Multiple-baseline assessments (three times per assessment phase) of speech fluency measures from sentence repetitions of untrained scripts were used to determine generalization of treatment effects. Nonoverlap of All Pairs was used to estimate effect size. The results showed significant direct and generalization of treatment effects for both treatments on all measures (speech initiation latency, speech event duration, speech rate and articulatio rate in AE2. Similar gains were seen in AE3 but there was no improvement in speech initiation latency following script training. In conclusion, both participants showed improvement on most of the measures following both treatment blocks. Verbal self-feedback may be a promising tool to improve speech production efficiency in nonfluent aphasia.