Helplessness under rescue rule BASE, k=3. CONTROL: Helplessness has a gap of only 0.105, below the 0.25 step cap, so the STRICT rule already works here -- it was never blocked. 88.8 % of clips score at or below zero on this emotion and the largest gap on its normalised axis is 0.105 (narrower than the 0.25 step cap). This rule found 19,693 chains over 40,000 tracks; the strict rule found 19,693 at k=3.
BASE changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25. What it costs: Nothing -- this is the strict rule everything else is measured against. 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.76 — a total move of +0.35.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.35 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 100–150 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 0.663 before conversion and 0.677 after — it rose by 0.014. Neighbour-to-neighbour the worst pair went 0.663 → 0.664. (The earlier render, with segment 1 left raw, scores 0.502 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, Helplessness moved +0.353 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 2.88 → 3.15 (+0.28) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 100–150 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 0.911 before conversion and 0.917 after — it rose by 0.006. Neighbour-to-neighbour the worst pair went 0.912 → 0.917. (The earlier render, with segment 1 left raw, scores 0.843 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.
The re-scored move on Helplessness is +0.000 before and +0.000 after, but the before-value is too close to zero for a retention ratio to mean anything on this chain.
Quality. Mean predicted overall quality across the segments went 3.13 → 3.24 (+0.11) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, 0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at 0.00.
On the corpus-wide percentile scale those become 0.41, 0.76, 0.86 — a total move of +0.46.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Helplessness, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Helplessness 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: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.014 before conversion and 0.347 after — it rose by 0.333. Neighbour-to-neighbour the worst pair went 0.361 → 0.512. (The earlier render, with segment 1 left raw, scores 0.283 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, Helplessness moved +0.456 in the original and +0.510 after conversion — 112 % of the delta retained, i.e. the move came out slightly larger after conversion than before.
Quality. Mean predicted overall quality across the segments went 2.85 → 3.06 (+0.21) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.867 before conversion and 0.786 after — it fell by 0.081. Neighbour-to-neighbour the worst pair went 0.815 → 0.682. (The earlier render, with segment 1 left raw, scores 0.705 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.
The re-scored move on Helplessness is +0.000 before and +0.437 after, but the before-value is too close to zero for a retention ratio to mean anything on this chain.
Quality. Mean predicted overall quality across the segments went 3.03 → 3.24 (+0.21) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.846 before conversion and 0.842 after — it fell by 0.004. Neighbour-to-neighbour the worst pair went 0.821 → 0.847. (The earlier render, with segment 1 left raw, scores 0.785 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.
The re-scored move on Helplessness is +0.000 before and +0.000 after, but the before-value is too close to zero for a retention ratio to mean anything on this chain.
Quality. Mean predicted overall quality across the segments went 2.94 → 3.20 (+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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, 0.11. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at 0.11.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.91 — a total move of +0.50.
That is why the strict rule cannot build this chain. Roughly 90 % of the corpus scores exactly zero on Helplessness, and tied values all collapse onto one point (about 0.45). The clips that genuinely carry Helplessness 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: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.882 before conversion and 0.900 after — it rose by 0.018. Neighbour-to-neighbour the worst pair went 0.882 → 0.900. (The earlier render, with segment 1 left raw, scores 0.845 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, Helplessness moved +0.496 in the original and +0.459 after conversion — 93 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.90 → 3.17 (+0.27) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.76 — a total move of +0.35.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.35 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.387 before conversion and 0.640 after — it rose by 0.253. Neighbour-to-neighbour the worst pair went 0.387 → 0.640. (The earlier render, with segment 1 left raw, scores 0.574 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.
The emotional move did not survive. Re-scored end to end, Helplessness moved +0.353 in the original and -0.353 after conversion — it changed direction. On this chain the corrected audio is not an improvement.
Quality. Mean predicted overall quality across the segments went 2.81 → 3.11 (+0.30) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.07, 0.41, 0.41 — a total move of +0.34.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.34 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 100–150 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 0.862 before conversion and 0.858 after — it fell by 0.004. Neighbour-to-neighbour the worst pair went 0.819 → 0.873. (The earlier render, with segment 1 left raw, scores 0.788 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, Helplessness moved +0.341 in the original and +0.341 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.82 → 3.05 (+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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.76 — a total move of +0.35.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.35 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.649 before conversion and 0.649 after — it rose by 0.000. Neighbour-to-neighbour the worst pair went 0.588 → 0.472. (The earlier render, with segment 1 left raw, scores 0.453 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, Helplessness moved +0.353 in the original and +0.353 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.60 → 2.92 (+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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.07, 0.41, 0.41 — a total move of +0.34.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.34 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 100–150 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 0.751 before conversion and 0.835 after — it rose by 0.084. Neighbour-to-neighbour the worst pair went 0.842 → 0.866. (The earlier render, with segment 1 left raw, scores 0.761 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, Helplessness moved +0.341 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 2.98 → 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 100–150 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 0.822 before conversion and 0.836 after — it rose by 0.013. Neighbour-to-neighbour the worst pair went 0.833 → 0.855. (The earlier render, with segment 1 left raw, scores 0.765 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.
The re-scored move on Helplessness is +0.000 before and +0.000 after, but the before-value is too close to zero for a retention ratio to mean anything on this chain.
Quality. Mean predicted overall quality across the segments went 2.78 → 3.16 (+0.39) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.818 before conversion and 0.824 after — it rose by 0.006. Neighbour-to-neighbour the worst pair went 0.818 → 0.855. (The earlier render, with segment 1 left raw, scores 0.761 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.
The re-scored move on Helplessness is +0.000 before and +0.000 after, but the before-value is too close to zero for a retention ratio to mean anything on this chain.
Quality. Mean predicted overall quality across the segments went 3.17 → 3.42 (+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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.221 before conversion and 0.808 after — it rose by 0.587. Neighbour-to-neighbour the worst pair went 0.157 → 0.784. (The earlier render, with segment 1 left raw, scores 0.768 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, Helplessness moved +0.353 in the original and +0.532 after conversion — 151 % of the delta retained, i.e. the move came out slightly larger after conversion than before.
Quality. Mean predicted overall quality across the segments went 3.10 → 3.29 (+0.19) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.818 before conversion and 0.866 after — it rose by 0.049. Neighbour-to-neighbour the worst pair went 0.871 → 0.897. (The earlier render, with segment 1 left raw, scores 0.739 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.
The re-scored move on Helplessness is +0.000 before and +0.353 after, but the before-value is too close to zero for a retention ratio to mean anything on this chain.
Quality. Mean predicted overall quality across the segments went 3.02 → 3.29 (+0.27) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.237 before conversion and 0.656 after — it rose by 0.418. Neighbour-to-neighbour the worst pair went 0.280 → 0.484. (The earlier render, with segment 1 left raw, scores 0.545 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.
The re-scored move on Helplessness is +0.000 before and -0.341 after, but the before-value is too close to zero for a retention ratio to mean anything on this chain.
Quality. Mean predicted overall quality across the segments went 2.77 → 3.16 (+0.39) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.76 — a total move of +0.35.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.35 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.755 before conversion and 0.699 after — it fell by 0.056. Neighbour-to-neighbour the worst pair went 0.755 → 0.691. (The earlier render, with segment 1 left raw, scores 0.703 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.
The emotional move did not survive. Re-scored end to end, Helplessness moved +0.353 in the original and -0.119 after conversion — it changed direction. On this chain the corrected audio is not an improvement.
Quality. Mean predicted overall quality across the segments went 2.68 → 2.97 (+0.29) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.41 — a total move of +0.00.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.00 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 100–150 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 0.034 before conversion and 0.543 after — it rose by 0.509. Neighbour-to-neighbour the worst pair went 0.034 → 0.522. (The earlier render, with segment 1 left raw, scores 0.467 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.
The re-scored move on Helplessness is +0.000 before and +0.353 after, but the before-value is too close to zero for a retention ratio to mean anything on this chain.
Quality. Mean predicted overall quality across the segments went 2.57 → 2.99 (+0.42) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.76 — a total move of +0.35.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.35 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.579 before conversion and 0.713 after — it rose by 0.134. Neighbour-to-neighbour the worst pair went 0.579 → 0.748. (The earlier render, with segment 1 left raw, scores 0.645 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, Helplessness moved +0.353 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 2.85 → 3.03 (+0.17) 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 BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.76 — a total move of +0.35.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.35 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 100–150 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 0.747 before conversion and 0.762 after — it rose by 0.015. Neighbour-to-neighbour the worst pair went 0.747 → 0.817. (The earlier render, with segment 1 left raw, scores 0.788 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, Helplessness moved +0.353 in the original and +0.353 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 3.13 → 3.26 (+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.
This is not a strict-rule trajectory. It comes from rescue rule BASE, which exists because the strict rule returns nothing at all for Helplessness.
The raw scorer output across the chain is -0.00, -0.00, -0.00. In the first clip the scorer found no Helplessness whatsoever (-0.00); by the last it is at -0.00.
On the corpus-wide percentile scale those become 0.41, 0.41, 0.76 — a total move of +0.35.
That is why the strict rule cannot build this chain — but for the opposite reason to the jump case. Every clip here already carries some Helplessness, so they all land in the crowded top tenth of the corpus-wide ranking, where roughly 90 % of clips scoring zero have consumed everything below. Measured that way the chain moves only 0.35 in total, far short of the 0.25 the strict rule demands — even though the raw scores clearly rise.
Re-ranked among the Helplessness-bearing clips only — which is exactly what this rule does — the same clips read 0.00, 0.00, 0.00, a move of +0.00, which is usable again.
What this rule changes: The strict rule, unchanged, as a control. Rank-normalise the emotion over the whole corpus, then require the chain to rise by at least 0.25 end-to-end with every consecutive step at most 0.25.
What it costs: Nothing -- this is the strict rule everything else is measured against.
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 3 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 0.837 before conversion and 0.854 after — it rose by 0.017. Neighbour-to-neighbour the worst pair went 0.837 → 0.820. (The earlier render, with segment 1 left raw, scores 0.675 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, Helplessness moved +0.353 in the original and +0.353 after conversion — 100 % of the delta retained, which is essentially all of it.
Quality. Mean predicted overall quality across the segments went 2.84 → 3.20 (+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.