The potential of acoustic signatures to be used for State-of-Charge (SoC) estimation is demonstrated using artificial neural network regression models. This approach represents a streamlined method of processing the entire acoustic waveform instead of performing manual, and often arbitrary, waveform peak selection. For applications where computational economy is prioritised, simple metrics of statistical significance are used to formally identify the most informative waveform features. These alone can be exploited for SoC inference. It is further shown that signal portions representing both early and late interfacial reflections can correlate highly with the SoC and be of predictive value, challenging the more common peak selection methods which focus on the latter. Although later echoes represent greater through-thickness coverage, and are intuitively more information-rich, their presence is not guaranteed. Holistic waveform treatment offers a more robust approach to correlating acoustic signatures to electrochemical states. It is further demonstrated that transformation into the frequency domain can reduce the dimensionality of the problem significantly, while also improving the estimation accuracy. Most importantly, it is shown that acoustic signatures can be used as sole model inputs to produce highly accurate SoC estimates, without any complementary voltage information. This makes the method suitable for applications where redundancy and diversification of SoC estimation approaches is needed. Data is obtained experimentally from a 210 mAh LiCoO₂/graphite pouch cell. Mean estimation errors as low as 0.75% are achieved on a SoC scale of 0–100%.