Note 19:

Allophones of /b/, /d/ and /g/

Allophones of /b/: although /b/ is a single phoneme, its exact realisation depends on context: it is a voiced stop after a pause or a consonant that isn’t a written s. Phonetically, this is [b].
But in other contexts, like between vowels and after s (see note ³, item ¹), there is a tendency to make the closure incomplete, and introduce some weak friction. Phonetically, this is [B]. Example: Lisboa. But this isn’t always done, or not always to the same extent: Lisboa.
And to show that this isn’t a peculiarity of just this one singer, here are two more examples by another singer, in which both times you first hear something that is almost [b], and then in the same word a full [B]: Sabem ^©, Sabem ^©.
See also the "afago" sound samples ^©, for a similar phonetic alternation.
Cf. Spanish, where the same thing happens with written g, d and b/v, and cf. Portuguese g and d. But contrary to Spanish, in Portuguese /b/ and /v/ are distinct phonemes.
The sample vai o bem fugindo ^© shows the difference between labiodental /v/ (in "vai") and bilabial [B] (in "o bem").
Allophones of /d/: although /d/ is a single phoneme, its exact realisation depends on context: it is a voiced stop after a pause or a consonant that isn’t a written s. Phonetically, this is [d]. It is a dental [d_d], unlike the alveolar [d] sound found in English.
But in other contexts, like between vowels and after s (see note ³, item ¹), there is a tendency to make the closure incomplete, and introduce some weak friction. Phonetically, this is [D]. It is very similar to th in English "the", and to the sound of the delta in modern Greek.
Cf. Spanish, where the same allophonic change occurs with written g, d and b/v, and cf. Portuguese g and b.
Allophones of /g/: although /g/ is a single phoneme, its exact realisation depends on context: it is a voiced stop in positions lg, ng, rg, Bg ¹. Phonetically, this is [g].
But in other contexts, like between vowels (note ³, item ¹), there is a tendency to make the closure incomplete, and introduce some weak friction. Phonetically, this is [G]. It doesn’t happen consistently (in the sample it is very clear the last time, but much less so the previous time), and the friction is nowhere near that of a Dutch gg or an Arabic ghain.
Cf. Spanish, where the same thing happens with written g, d and b/v, and cf. Portuguese d and b.

The extent of the friction, and how much of the plosion is absent, varies even with the same speaker and in the same song. This becomes clear when listening to these two samples: the first one is more like [3fagu] ^©, and only the second one clearly has this friction element: [3faGu] ^©. See also the Lisboa sound samples.