Voicing is defined as the semi-periodic vibration of the vocal folds. Accordingly, you look to see if there is something that happens repeatedly at a reasonably low but not ridiculously low frequency, e.g. for a given speaker it might be in the range of 80-200 Hz, but higher or a wider range for another. The technique of autocorrelation can be used to detect such repeated patterns (and underlies computation of pitch). The idea of autocorrelation is to look at the shape of a piece of waveform, and find the most similar part that follows it (within an appropriate window). If there is enough of a semi-regular pattern, you have voicing. If you have creaky voice, this method may well not detect the glottal pulses. If you are not looking at fricatives, you can look for visible voicing pulses in the spectrogram.
You can also look at a spectrogram, looking for the "voice bar", which is high amplitude at a low frequency. With voiceless fricatives, you can get high amplitude from the turbulent airflow, but the frequency is up there above 1KHz: high amplitude in the realm of 200 Hz does not come from a noisy (aperiodic) source. The problem with voice bar as a detector is that with stops especially, the amplitude of the glottal pulses can be low enough that it's hard to detect. Also, if you are looking at spectral slices, you have to make sure that the low-frequency amplitude is enough higher than the amplitude at higher frequencies (this is an issue in case you intend some automated voicing detection and don't have ideal recordings).
Ultimately the various acoustic methods have reasonable but not total success, and the only way to determine if the vocal folds are vibrating is physiologically.