Awe under rescue rule S3, k=2. does the winning rule generalise? Awe has the widest gap (0.474). 95.1 % of clips score at or below zero on this emotion and the largest gap on its normalised axis is 0.474 (WIDER than the 0.25 step cap). This rule found 5,762 chains over 40,000 tracks; the strict rule found 0 at k=3.
S3 changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'. What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained. Full explanation →Interest has a median of 2.08 and is never zero, while Infatuation is zero on 87.7 % of clips — and the caption named an emotion whenever its raw score cleared an absolute 1.0. Interest therefore appeared in 94.8 % of captions and Sadness in almost none: the clause was reporting the scale of the head, not the emotion of the clip. An emotion is now named only when it lands in the top 10 % for that emotion, against a pooled tie-aware mid-rank ECDF over 132,833,726 utterances spanning every dataset and language. Interest now appears in 6.2 %, all 40 emotions occur, and a clip that is ordinary on all 40 says “no dominant emotion” rather than being forced to pick one (21.8 % of clips). This is the same scale the trajectory miner selects on, so the caption and the mining now refer to the same quantity: the mined target emotion is named in the final clip's caption on 73 % of chains, up from 46 %.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.31. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.31.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.923 before conversion and 0.909 after — it fell by 0.014. Neighbour-to-neighbour the worst pair went 0.923 → 0.909. (The earlier render, with segment 1 left raw, scores 0.818 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.523 in the original and +0.511 after conversion — 98 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 3.16 → 3.28 (+0.12) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.16. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.16.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.848 before conversion and 0.805 after — it fell by 0.043. Neighbour-to-neighbour the worst pair went 0.848 → 0.805. (The earlier render, with segment 1 left raw, scores 0.718 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.520 in the original and +0.521 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.89 → 3.14 (+0.25) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.10. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.10.
On the corpus-wide percentile scale those become 0.47, 0.99 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was -0.058 before conversion and 0.540 after — it rose by 0.599. Neighbour-to-neighbour the worst pair went -0.058 → 0.540. (The earlier render, with segment 1 left raw, scores 0.168 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.519 in the original and +0.515 after conversion — 99 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.65 → 2.80 (+0.15) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.32. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.32.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.822 before conversion and 0.725 after — it fell by 0.097. Neighbour-to-neighbour the worst pair went 0.822 → 0.725. (The earlier render, with segment 1 left raw, scores 0.597 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.523 in the original and +0.000 after conversion — 0 % of the delta retained, so a meaningful part of the trajectory was flattened.
Quality. Mean predicted overall quality across the segments went 3.08 → 3.22 (+0.14) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.16. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.16.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.262 before conversion and 0.747 after — it rose by 0.484. Neighbour-to-neighbour the worst pair went 0.262 → 0.747. (The earlier render, with segment 1 left raw, scores 0.682 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.520 in the original and +0.519 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.79 → 3.01 (+0.22) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.42. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.42.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.724 before conversion and 0.675 after — it fell by 0.049. Neighbour-to-neighbour the worst pair went 0.724 → 0.675. (The earlier render, with segment 1 left raw, scores 0.605 here.) This chain started 0.70-0.80 — close, but under the identity threshold, a band where the conversion is close to a wash on this measure.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.524 in the original and +0.512 after conversion — 98 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.97 → 3.23 (+0.25) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.19. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.19.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.068 before conversion and 0.693 after — it rose by 0.624. Neighbour-to-neighbour the worst pair went 0.068 → 0.693. (The earlier render, with segment 1 left raw, scores 0.703 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.521 in the original and +0.508 after conversion — 98 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.86 → 3.00 (+0.14) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.12. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.12.
On the corpus-wide percentile scale those become 0.47, 0.99 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.227 before conversion and 0.787 after — it rose by 0.559. Neighbour-to-neighbour the worst pair went 0.227 → 0.787. (The earlier render, with segment 1 left raw, scores 0.773 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.519 in the original and +0.515 after conversion — 99 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.94 → 3.06 (+0.12) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.02. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.02.
On the corpus-wide percentile scale those become 0.47, 0.99 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.681 before conversion and 0.742 after — it rose by 0.061. Neighbour-to-neighbour the worst pair went 0.681 → 0.742. (The earlier render, with segment 1 left raw, scores 0.553 here.) This chain started 0.50-0.70 — audibly different, a band where the conversion is close to a wash on this measure.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.517 in the original and +0.516 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.78 → 3.12 (+0.35) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.02. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.02.
On the corpus-wide percentile scale those become 0.47, 0.99 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.725 before conversion and 0.847 after — it rose by 0.122. Neighbour-to-neighbour the worst pair went 0.725 → 0.847. (The earlier render, with segment 1 left raw, scores 0.657 here.) This chain started 0.70-0.80 — close, but under the identity threshold, a band where the conversion is close to a wash on this measure.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.517 in the original and +0.017 after conversion — 3 % of the delta retained, so a meaningful part of the trajectory was flattened.
Quality. Mean predicted overall quality across the segments went 2.97 → 3.21 (+0.24) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.07. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.07.
On the corpus-wide percentile scale those become 0.47, 0.99 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.797 before conversion and 0.865 after — it rose by 0.069. Neighbour-to-neighbour the worst pair went 0.797 → 0.865. (The earlier render, with segment 1 left raw, scores 0.737 here.) This chain started 0.70-0.80 — close, but under the identity threshold, a band where the conversion is close to a wash on this measure.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.518 in the original and +0.516 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 3.13 → 3.33 (+0.20) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.16. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.16.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.831 before conversion and 0.774 after — it fell by 0.057. Neighbour-to-neighbour the worst pair went 0.831 → 0.774. (The earlier render, with segment 1 left raw, scores 0.706 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.520 in the original and +0.519 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.86 → 3.06 (+0.20) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.01. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.01.
On the corpus-wide percentile scale those become 0.47, 0.99 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.111 before conversion and 0.715 after — it rose by 0.604. Neighbour-to-neighbour the worst pair went 0.111 → 0.715. (The earlier render, with segment 1 left raw, scores 0.423 here.) This chain started below 0.50 — the segments really were different people, the band the conversion helps most: chains starting below 0.50 improve on this measure almost without exception.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.516 in the original and +0.524 after conversion — 101 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.71 → 2.97 (+0.26) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.49. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.49.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.809 before conversion and 0.834 after — it rose by 0.025. Neighbour-to-neighbour the worst pair went 0.809 → 0.834. (The earlier render, with segment 1 left raw, scores 0.819 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.524 in the original and +0.525 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.91 → 3.05 (+0.15) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.28. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.28.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.770 before conversion and 0.834 after — it rose by 0.064. Neighbour-to-neighbour the worst pair went 0.770 → 0.834. (The earlier render, with segment 1 left raw, scores 0.749 here.) This chain started 0.70-0.80 — close, but under the identity threshold, a band where the conversion is close to a wash on this measure.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.522 in the original and +0.000 after conversion — 0 % of the delta retained, so a meaningful part of the trajectory was flattened.
Quality. Mean predicted overall quality across the segments went 2.96 → 3.22 (+0.26) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.32. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.32.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.794 before conversion and 0.781 after — it fell by 0.013. Neighbour-to-neighbour the worst pair went 0.794 → 0.781. (The earlier render, with segment 1 left raw, scores 0.725 here.) This chain started 0.70-0.80 — close, but under the identity threshold, a band where the conversion is close to a wash on this measure.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.523 in the original and +0.524 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.66 → 2.91 (+0.26) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.32. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.32.
On the corpus-wide percentile scale those become 0.47, 1.00 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.921 before conversion and 0.955 after — it rose by 0.034. Neighbour-to-neighbour the worst pair went 0.921 → 0.955. (The earlier render, with segment 1 left raw, scores 0.848 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.523 in the original and +0.522 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.84 → 3.09 (+0.25) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.00. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.00.
On the corpus-wide percentile scale those become 0.47, 0.99 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.908 before conversion and 0.923 after — it rose by 0.015. Neighbour-to-neighbour the worst pair went 0.908 → 0.923. (The earlier render, with segment 1 left raw, scores 0.866 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.516 in the original and +0.489 after conversion — 95 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.88 → 3.24 (+0.36) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.06. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.06.
On the corpus-wide percentile scale those become 0.47, 0.99 — a total move of +0.52.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 150–150 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.824 before conversion and 0.793 after — it fell by 0.031. Neighbour-to-neighbour the worst pair went 0.824 → 0.793. (The earlier render, with segment 1 left raw, scores 0.673 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.518 in the original and +0.479 after conversion — 92 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.77 → 3.09 (+0.32) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.
This is not a strict-rule trajectory. It comes from rescue rule S3, which exists because the strict rule returns nothing at all for Awe.
The raw scorer output across the chain is -0.00, 1.56. In the first clip the scorer found no Awe whatsoever (-0.00); by the last it is at 1.56.
On the corpus-wide percentile scale those become 0.00, 1.00 — a total move of +1.00.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Awe, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Awe sit above 0.85. There is simply nothing in between to step onto, so the only move available is one jump far wider than the 0.25 per-step cap.
What this rule changes: Absent to present. The chain must start at or below 0.05 (the emotion is absent) and end at or above 1.0 (it is clearly present). The most literal reading of 'from not sad to sad'.
What it costs: Says nothing about the SHAPE of the path -- only that it begins absent and ends present. Any intermediate clip is unconstrained.
Same speaker? No similarity score is available here — the emolia clips in this chain are not covered by either speaker-embedding store. The chain therefore rests on the corpus's own speaker/track labelling, which is not the same as a measured check.
Voice consistency: these clips are separate recordings joined together, and no voice-similarity check could be run for this sample, so there is no measurement of how closely the voices match. You may hear the voice shift between segments. Voice conversion has not been applied yet in this build. A planned pass will re-render every segment onto the first segment's voice, which removes this effect entirely.
What was done to this chain. All 2 segments were re-synthesised with ChatterboxVC onto segment 1's voice — including segment 1 itself, converted with itself as the target — and then restored with SIDON. Converting the anchor too is what keeps the room and the reverb the same across the whole chain; leaving it raw put a change of acoustic at the first join. The words, timing and delivery still come from each original clip. The joins are 100–100 ms equal-power crossfades. The chain is normalised as one signal, so the loudness differences between segments are the ones the conversion produced, not a per-clip reset.
Did it unify the voice? On the 250-dimensional Speaker-wavLM-id verification embedding, the worst similarity between any segment and segment 1 was 0.945 before conversion and 0.943 after — it fell by 0.002. Neighbour-to-neighbour the worst pair went 0.945 → 0.943. (The earlier render, with segment 1 left raw, scores 0.916 here.) This chain started at or above 0.80 — already effectively one voice. Across the whole build that is the band where conversion tends to cost identity agreement rather than add it, and a hard identity cut would keep this chain without converting it at all. Judge it by ear against the original above.
Did the emotion survive? Re-scored end to end through the same emotion stack and the same corpus-percentile scale the chain was mined on, Awe moved +0.997 in the original and +0.029 after conversion — 3 % of the delta retained, so a meaningful part of the trajectory was flattened.
Quality. Mean predicted overall quality across the segments went 3.02 → 3.15 (+0.13) on the Empathic-Insight head. The corrected render is 48 kHz because SIDON outputs 48 kHz; the original is the 24 kHz source. Some of what you hear as “cleaner” is that bandwidth, not the conversion.