This paper investigates the impact of managers’ speech sentiment on stock returns using a machine learning technique to analyze the acoustic features of earnings conference calls. The paper contributes to the existing literature by proposing a novel method to measure the emotional state of managers and by using a comprehensive dataset of over 60,000 earnings call observations from 2018 to 2022. The paper finds that managers’ speech sentiment has a significant and persistent effect on stock returns, both at the firm level and at the industry level. The paper also examines the channels through which speech sentiment affects stock returns, such as investor attention, analyst forecasts, and media coverage. The paper provides evidence that speech sentiment is an important source of information for investors and market participants and that it can complement or substitute other textual sentiment measures.