[SSML] Wrong settings used for speech synthesis

[roehrt]

Describe the bug
Mimic3 ignores <prosody> settings and instead applies the settings of the last closed <prosody> block instead.

To Reproduce
mimic3 '<prosody rate="200%">This should be spoken fast but is not.</prosody><prosody volume="30%">This should be a bit quieter but is actually spoken faster</prosody>' --ssml | aplay

Expected behavior
Mimic3 should speak the first sentence fast and the second one with lowered volume.

Environment
 - Device type: desktop
 - OS: Ubuntu 22.04

Source of actual behavior

mimic3/mimic3_tts/tts.py

Lines 470 to 501 in be72c18

	def end_utterance(self) -> typing.Iterable[BaseResult]:
	last_settings: typing.Optional[Mimic3Settings] = None
	sent_phonemes: PHONEMES_LIST_TYPE = []
	for result in self._results:
	if isinstance(result, Mimic3Phonemes):
	if result.is_utterance:
	# Utterance boundary
	if (
	sent_phonemes
	and (last_settings is not None)
	and (result.current_settings != last_settings)
	):
	# Not compatible with existing utterance.
	# Need to speak previous utterance first.
	yield self._speak_sentence_phonemes(
	sent_phonemes, settings=last_settings
	)
	sent_phonemes.clear()
	# Current utterance
	sent_phonemes.extend(result.phonemes)
	if sent_phonemes:
	yield self._speak_sentence_phonemes(
	sent_phonemes, settings=last_settings
	)
	sent_phonemes.clear()
	else:
	# Continue until utterance boundary
	sent_phonemes.extend(result.phonemes)
	last_settings = result.current_settings

[stephenrt42]

Try this;
<speak> <prosody rate="200%"><s>This should be spoken fast but is not.</s></prosody><prosody volume="30%"><s>This should be a bit quieter but is actually spoken faster</s></prosody> </speak>