Accurate respiratory monitoring is of great significance in assessing and analyzing physical health, and preventing respiratory diseases. The recently emerged wearable respiratory sensors are confronted with the challenges such as complex fabrication processes, limited accuracy, and stringent wearing requirements. An optical fiber sensor for accurate human respiratory monitoring is proposed and experimentally verified. The sensor head is composed of a piece of seven core fiber sandwiched between two single-mode fibers by two fiber bitapers, which is embedded in a textile sheet and freely worn on the upper body. An efficient signal demodulation system is set up to acquire the respiratory signal, while Fourier transform (FFT) and short-time Fourier transform (STFT) methods are used to analyze the measured signal. Six volunteers are invited to perform the respiratory experiment, and the experimental results demonstrate that the sensor can accurately detect and distinguish respiratory signals under different humans, different states (normal, slow, fast), different body parts (abdomen, chest, back), different postures (standing, sitting, lying), and irregular respiration. The Pearson correlation coefficient of the sensor is higher than 0.9, which is consistent with commercial respiratory sensor. Meanwhile, the instability of the sensor is 0.003 Hz for the same volunteer in 6 months. The sensor has the advantages of high sensitivity, good stability and wearing comfort, showing good potential in healthcare applications.