Cough is a characteristic symptom of tuberculosis, is the main cause of transmission, and is used to assess treatment response. We aimed to identify the best measure of cough severity and characterize changes during initial tuberculosis therapy. We conducted a prospective cohort of recently diagnosed ambulatory adult patients with pulmonary tuberculosis in two tertiary hospitals in Lima, Peru. Pre-treatment and five times during the first two months of treatment, a vibrometer was used to capture 4-hour recordings of involuntary cough. A total of 358 recordings from 69 participants were analyzed using a computer algorithm. Total time spent coughing (seconds per hour) was a better predictor of microbiologic indicators of disease severity and treatment response than the frequency of cough episodes or cough power. Patients with prior tuberculosis tended to cough more than patients without prior tuberculosis, and patients with tuberculosis and diabetes coughed more than patients without diabetes co-morbidity. Cough characteristics were similar regardless of HIV co-infection and for drug-susceptible versus drug-resistant tuberculosis. Tuberculosis treatment response may be meaningfully assessed by objectively monitoring the time spent coughing. This measure demonstrated that cough was increased in patients with TB recurrence or co-morbid diabetes, but not because of drug resistance or HIV co-infection.