### emitter-follower output stage

You have a great answer (I'll +1 it, shortly) exposing the problems with your output stage (emitter follower.) It's only active in one quadrant and therefore depends entirely upon a resistor (your \$47\:\Omega\$) for the other quadrant. As shown in that answer, if you match it up with your nameplate value for the speaker, then it can be made to perform better. But it will ***never*** be an active 2-quadrant output stage and as ***unawriter*** points out, the total dissipation reaches \$28\:\text{W}\$ to deliver \$1\:\text{W}\$.

Here's a picture of an old schematic template I put together almost a decade ago, which allows me to specify some details and then automatically generates the values to run the test (***Note the title?***):

[![enter image description here][1]][1]

The `speaker` gets \$1.0\:\text{W}\$. For this, the \$5.6\:\Omega\$ emitter resistor dissipates \$19.3\:\text{W}\$ and the BJT dissipates \$7.0\:\text{W}\$. This \$26.3\:\text{W}\$ total is surprisingly close to what ***unawriter*** wrote.

If I lower the **VCEMIN** parameter to \$1\:\text{V}\$ the emitter resistor goes to \$6.5\:\Omega\$ and this total circuit dissipation drops to about \$24.7\:\text{W}\$. Not much of a difference. Hopefully, you get the idea.

You can take it as gospel that this topology is not so good. Don't do it unless you are making **ElectroBOOM**-like YouTube videos.

### one class-A 2-quadrant approach

The following is a 2-quadrant design class-A design where I've set the quiescent current to a scrape-by level:

[![enter image description here][2]][2]

Here, the upper quadrant \$Q_1\$ dissipates \$4.3\:\text{W}\$ and the lower quadrant \$Q_2\$ dissipates \$3.6\:\text{W}\$ for a circuit total of \$7.9\:\text{W}\$.

Note the significant improvement by switching out a passive resistor in the 1-quadrant design, replacing it with an active component in the 2-quadrant design?

### another class-A 2-quadrant approach

Another design based upon something I wrote [here](https://electronics.stackexchange.com/a/695920), recently, might look like this:

[![enter image description here][3]][3]

*(I used a lower \$V_{_\text{CC}}\$ since that was all that was needed here.)*

The upper quadrant \$Q_1\$ dissipates \$3.5\:\text{W}\$ and the lower quadrant \$Q_2\$ dissipates \$3.0\:\text{W}\$ for a circuit total of \$6.5\:\text{W}\$.

### summary

The above makes some comparisons.

Both examples of 2-quadrant designs use global NFB to control the gain. So, unlike your approach, these provide stable gain against device variations and operating temperatures. They will also have much better THD and waste about \$3\times\$ less power.

But class-AB does much better, dissipating perhaps only a little more than the speaker itself. It introduces the possibility of cross-over distortion, which can be mitigated by running it a little 'hotter' and wasting a little more dissipation. But the global NFB covers many ills. So they are worth understanding and pursuing. (Especially when you are talking about delivering \$1\:\text{W}\$ and more into a speaker load.)

 [1]: https://i.sstatic.net/4TGdj.png
 [2]: https://i.sstatic.net/ehHu8.png
 [3]: https://i.sstatic.net/CxDyc.png
 [4]: https://i.sstatic.net/Jm3Cl.png