fixed align() missing nonspeech

-fixed `align()` not predicting nonspeech timings for the new section after it skips a nonspeech section due to `nonspeech_skip`; this can lead to missing nonspeech timings if next audio section does not completely overlap the new section (such as in #319 (comment))

Author: jianfch

stable_whisper/alignment.py

@@ -422,6 +422,7 @@ def redo_words(_idx: int = None):
                             audio_segment = audio.next_chunk(seek_sample, N_SAMPLES)
                             if audio_segment is None:
                                 break
+                            nonspeech_predictor.predict(audio=audio_segment, offset=time_offset)
                             if len(nonspeech_starts) > 1:
                                 new_sample_count = round((nonspeech_starts[1] - nonspeech_ends[0]) * SAMPLE_RATE)
                             else: