TOT is self-biased with negative DC feedback using R11=100k between C-B and emitter R2 Bypassed for high AC gain. Since T2 has a grounded emitter it also has high gain and inverting thus C2 is "positive AC feedback with gain >>1) which automatically makes it an oscillator. This type of oscillator is called a Relaxation type sawtooth type that uses the positive ramp from T1 cutoff with a negative base-emitter voltage. T1 Vb is [pulled up R1 until Vbe conducts then T2 rapidly saturates with a step that turns off T1 and thus truncates it to a very narrow pulse with a ramp time of R1*C2 = 1 ms approx. or 1 kHz.
The diagram shows the oscillation of one cycle of a sine wave but this is false.
It's just a <1us pulse every 1 ms.
The author just wanted to show (poorly) that there is a signal inversion in the collector output from the base (Common Emitter) somehow with the polarity inversion of a sine wave, even though I know it won't be a sine wave. Just remember that ALL common emitter amplifiers are inverting
and two stages in the feedback to base loop make it positive feedback.
So C2 is continually being charged by R1 pullup and pulled down with a step by T2 to discharge the ramp and then cut the pulse short to start another ramp with all the transistors OFF. The speaker is thus OFF, except for narrow 1kHz spikes ( that will sound like a buzz when a transistor is inserted in the socket.
It won't matter if the hFE is 5 to > 100 as the circuit will oscillate by the effects only of R1C2 to the sound of the speaker. But if the NPN C-E is reversed the hFE will be essentially about 1 and will not oscillate as there is sufficient attenuation in R4 to reduce the gain of T2.
The narrow pulse also makes the battery power consumption very small and zero with no transistor in. < 9V/ 100k = 90 uA during the ramp but several watts peak for a < 1us during the pulse.
This is the same circuit simulated in your browser.