I don't have personal experience with CI, but have researched it extensively as a potential fix for my hearing loss. From what I understand, the number of electrodes or links to neural tissue is the limiting factor. With something like NeuraLink with 1024 electrodes, the resolution of the sound signal can be much higher. For something like early cochlear implants, with 8 or 12 channels, the signal is going to be very digitized and artificial sounding; you have to heavily optimize for a particular type or modality of sound, and that's usually speech. That means all sorts of nuance things like music and voices get lost in compression, or filtered out entirely.
Cochlear implants are essentially BCI implants, taking the place of the cochlea in signaling via neural tissue.
To completely replicate natural sound, you'd likely need somewhere between 15,000 to 30,000 electrodes. It's not linear, however, and 8-12 electrodes might get you to sound that is about 25% of normal, and 1024 will get you to 85-90% normal. Full fidelity of sound, or even better, will be possible once we get implants working with many tens of thousands of electrodes. People will have senses that far exceed biological human limitations.
One neat thing with all of this is that due to plasticity, any connections on the neocortex can be trained to behave as if they're wired to any sensory organ; there aren't any hard limits on where an implant has to be connected. If you had an implant with 50k electrodes, half of them could be dedicated to sound, and the other half to sci-fi level possibilities like BCI mouse and keyboard control, simple virtual displays through modified sight, secondary audio channels, North sense, radar, electromagnetic signals, or immersion tweaks that modulate proprioceptive signaling.
1-500k would allow for convincing replication of normal sight, with the obvious advantage that with everything being digital, you'd be able to process your vision in software (Please watch this ad before waking! Skip in 10...).
With a million electrodes, you could get into convincing totally immersive full sensory simulation. There would be some resolution issues, initially, but we're some materials science, software design, and engineering problems away from full Matrix style simulations. 1 sq cm of neocortex is all you'd need for access to 1 million neurons - things are pretty densely packed, and all the neurons we need to access live on the outer surface of the brain.
Things are gonna start improving and the rate will accelerate, so hopefully we start seeing radical doublings of cochlear implant and other BCI capabilities in the near future.
TLDR; as much of normal hearing as possible is compressed down to around 100hz over 8-12 electrodes in a cochlear implant. This results in significant quality degradation compared to normal hearing, but it can be a huge boon to someone who is totally or profoundly deaf. Implant technology is experiencing a boom, and we're going to see a period of Moore's law like scaling of electrodes until implants reach parity with the rest of our computing technology.