We investigate structures of acoustic sources in a Mj = 0.9 high-speed subsonic isother-mal turbulent jet using input-output analysis. The dynamics of small perturbations about a Reynolds-averaged Navier-Stokes solutions (RANS) base flow are governed by the linearized Navier-Stokes (LNS) equations. The Ffowcs Williams-Hawkings (FW-H) method is coupled with linear input-output analysis to predict far-field acoustics from near-field flow data. To distill acoustically relevant aerodynamic sources, input forcings are restricted to the region containing high turbulent kinetic energy (TKE). For outputs, we specify a preferential eigen-direction so that we can obtain input forcings radiate sound at a designated direction only. As in the authors’ previous work on input-output analysis of turbulent jet noise radiating into full far-field, we recover non-compact wavepackets at any radiation angle, supporting one coherent source mechanism of turbulent jet noise. At lower radiation angle than the peak noise angle, the input mode represents jitter or decoherence of the optimal input mode for full-arc outputs. Wavepackets radiating sound at large angles, on the other hand, are more compact compared with those corresponding to low angles and thus more efficient. Furthermore, by repeating our analysis over a range of frequencies, we show that more degrees of freedom are needed in terms of the azimuthal wavenumber as radiation angle rotates towards the sideline direction. The acoustic spectra appear broadened by considering higher azimuthal wavenumber modes as radiation angle increases at high frequencies.