More often than not, the noise is glaringly evident. Regardless, we users are required to provide a noise profile, which -being a sample- is by definition inaccurate.
The breathing of the speaker or parts of his speech are included by us, and are therefore considered "noise". The dreaded "tinkle bells" (*) ensue.
The "Sound Cleaner II" application for example does a remarkable job at discerning what is noise and what is "show".
(*) Variously know as "wine glass", "metallic bubbles", "robotic", etc.
Retrieving data ...